Natural language interfaces to structured databases are becoming increasingly common, largely due to advances in large language models (LLMs) that enable users to query data using conversational input rather than formal query languages such as SQL. While this paradigm significantly improves usability and accessibility, it introduces new security risks, particularly the amplification of SQL injection vulnerabilities through the prompt-to-SQL translation process. Malicious users can exploit these mechanisms by crafting adversarial prompts that manipulate model behavior and generate unsafe queries. In this work, we propose a multi-layered security framework designed to detect and mitigate LLM-mediated SQL injection attacks. The framework integrates a front-end security shield for prompt sanitization, an advanced threat detection model for behavioral and semantic anomaly identification, and a signature-based control layer for known attack patterns. We evaluate the proposed framework under diverse and realistic attack scenarios, including prompt injection, obfuscated SQL payloads, and context-manipulation attacks. To ensure robustness, we generate and curate a comprehensive benchmark dataset of adversarial prompts and assess performance across a fine-tuned LLM configuration. Experimental results demonstrate that the proposed approach achieves high detection accuracy while maintaining low false-positive rates, significantly improving the secure deployment of LLM-powered database applications.

Assessing code automatically is a significant challenge in distance learning, especially in large online courses with limited teaching resources. Although auto-gradable programming exercises address scalability, creating enough high-quality exer-cises-particularly designing comprehensive unit tests-remains time-consuming and labor-intensive. To address this, we introduce a GPT-based feature that automates unit test generation for customized exercises. With a single button press, instructors can adapt existing exercises to meet specific teaching objectives while preserving auto-gradability. The AI-generated tests comprehensively cover potential edge cases that might otherwise be overlooked, thus reducing the need for manual oversight. An empirical evaluation with eight experienced educators showed these tests to be both thorough and time-efficient, achieving an average System Usability Scale (SUS) score of 81.79. Participants, who reported intermediate to advanced proficiency in designing manual unit tests and intermediate familiarity with AI tools like ChatGPT, praised the feature's ease of use and seamless workflow integration. Their combined expertise in teaching, coding, and AI-informed course development allowed them to provide insightful feedback on the practicality and reliability of our GPT-based solution. Our study includes a small participant pool ($\mathrm{n}=8$) and primarily focuses on Python, a language wellsupported by GPT. Future research will involve expanding the participant group, exploring additional programming languages, and assessing long-term tool performance and adaptability in diverse educational contexts. By harnessing GPT's language modeling capabilities, our approach addresses the gap between generic, limited-coverage test generation and the need for robust, domain-specific tests. Early reports from participants suggest that specialized exercises-such as those involving advanced data structures-can also benefit from automated unit test generation, though further evaluation is necessary. By leveraging artificial intelligence, this method streamlines exercise customization and enhances the overall usability and effectiveness of programming education tools. It has the potential to revolutionize auto-gradable exercise creation at scale, empowering educators to deliver high-quality instruction while tackling both the technical and pedagogical challenges in programming education.

Massive Open Online Courses (MOOCs) provide access to everyone to learn quality courses flexibly. However, a low engagement rate is still a challenge for MOOCs, especially in a long course series that comprehensively addresses a topic. Interactive Elements (IE) have been proposed as a solution to increase motivation and raise learners' engagement levels. This study addresses the gap in the lack of research on the utilization of IE within the scope of the series of courses. By examining the course series and the use of IE, the engagement was then measured and compared with the extensive use of the IE. Subsequently, the feedback from the user survey was analyzed using latent Dirichlet Allocation (LDA) topic modeling to obtain data on theme congruence with IE utilization in the course series. The result shows evidence of user satisfaction with various IEs integrated into the course series, which serves as a mechanism to increase learners' engagement levels. However, the use of IE is not only a deciding factor since the problem of how IE is being presented and other technical issues can also hinder the engagement level of the learners. This study provides data-driven insights into the strategic integration of interactive elements in MOOC course series, offering recommendations for effective content design to optimize learner motivation and engagement.

Large Vision-Language Models (LVLMs) with discrete image tokenizers unify multimodal representations by encoding visual inputs into a finite set of tokens. Despite their effectiveness, we find that these models still hallucinate non-existent objects. We hypothesize that one reason is due to visual priors induced during training: when certain image tokens frequently co-occur in the same spatial regions and represent shared objects, they become strongly associated with the verbalizations of those objects. As a result, the model may hallucinate by evoking visually absent tokens that often co-occur with present ones. To test this assumption, we construct a co-occurrence graph of image tokens using a segmentation dataset and employ a Graph Neural Network (GNN) with contrastive learning followed by a clustering method to group tokens that frequently co-occur in similar visual contexts. We find that hallucinations predominantly correspond to clusters whose tokens dominate the input, and more specifically, that the visually absent tokens in those clusters show much higher correlation with hallucinated objects compared to tokens present in the image. Based on this observation, we propose a hallucination mitigation method that suppresses the influence of visually absent tokens by modifying latent image embeddings during generation. Experiments show our method reduces hallucinations while preserving expressivity.

In the context of programming education, so-called auto-graders allow learners to receive automated feedback on their submissions. Because assessing learners' code typically involves executing the learners' untrusted code, this commonly used mechanism poses a significant security risk for these systems. Since auto-graders are mostly employed in the context of large-scale learning environments, such as universities or Massive Open Online Courses (MOOCs), security considerations are especially important. In this paper, we first introduce our auto-grader CodeOcean, which is regularly used in MOOCs with thousands of active learners, and in university contexts. As the execution of untrusted code can entail severe security implications, ensuring that the application contains no security vulnerabilities is essential. Hence, we partnered with a security consultancy to assess our auto-grader system landscape through a professional penetration test. This work presents the findings and countermeasures resulting from the performed security analysis for CodeOcean. We contextualize overarching enhancements for three main categories of threat vectors to auto-grader systems. Implementing these in any auto-grader system can improve the security and prevent learners from manipulating the assessment of their code. We also discuss the potential consequences of hardening an auto-grader, such as a reduced system performance. Therewith, we provide valuable recommendations for educators, researchers, and system designers to improve the security of auto-graders in the future, supporting their usage in even larger settings or in the context of exams.

Generalized Class Discovery (GCD) seeks to identify both known and unknown categories within an unlabeled dataset, utilizing the knowledge from a labeled dataset of known classes. Existing research implicitly/explicitly assumes that the frequency of occurrence for each category, whether known or unknown, is approximately the same in the unlabeled data. However, real-world scenarios often exhibit a long-tailed distribution of visual classes, where known or common classes appear more frequently than unknown or rare ones. Addressing this discrepancy, we introduce a new challenge: Imbalanced Generalized Category Discovery (IGCD), which deals with an imbalanced distribution in unlabeled data, favoring known over unknown classes. To tackle this, we propose a novel Optimal Transport-based Expectation Maximization framework for Imbalanced Generalized Category Discovery (Otem-IGCD) by aligning the marginal class prior distribution. Otem-IGCD also incorporates a systematic mechanism for estimating the imbalanced class prior distribution under the GCD setup. Our comprehensive experiments reveal that Otem-IGCD surpasses previous state-of-the-art GCD methods by achieving an improvement of approximately 2 - 4% on CIFAR100 and 15 - 19% on ImageNet-100, indicating its superior effectiveness in solving the Imbalanced GCD problem.

Optimizing General Matrix Multiplication (GEMM) on GPU platforms has become increasingly important due to the scaling demands of modern deep neural network research. While substantial progress has been made in accelerating high-precision GEMM, optimizing lower-bit GEMM remains an open problem. The CUTLASS library offers highly optimized low-bit GEMM based on tensor cores, but performance varies significantly with tile and pipeline settings across different GPUs. We introduce a novel auto-tuning framework for low-bit CUTLASS GEMM that employs a neural network model to predict optimal GEMM template parameters for target GPUs. This model was trained on a synthetic dataset featuring various matrix sizes from different Ampere GPUs and evaluated on these GPUs. In the test dataset, our method achieved an accuracy of up to 92.9%. Real-time evaluations of low-bit data types on the A100 GPU demonstrated speedups of up to 2.03× for GEMM and 1.44× for the linear layer compared to the default templates.

With a rising amount of highly successful educational content on major video platforms, science communication (SciComm) can be considered mainstream. Although the success in terms of social media metrics (e.g. followers and watch time) is undoubtedly given, the learning mechanisms of these production styles is under-researched. Through a between-subject-design of 980 adult learners in a MOOC about data science, this study analyzes how much of a difference four popular SciComm production styles about relational databases make in regard to perceived quality, learning success and technical user behavior. Testing the isolated effect showed no statistical difference in the grand scheme of things. Additionally, a multivariate regression model, estimating the overall course points with robust standard errors showed six significant variables: The time spend with the material and the number of exercise submissions are particular noteworthy. Based on our results, an underlying (video) script is more relevant than the actual production style. Prioritizing the preparation of this material instead following a specific, pre-existing video production style is recommended.

Massive Open Online Courses (MOOCs) are a popular form of online education that often attracts a huge and heterogeneous group of learners with diverse interests and backgrounds. However, most MOOCs follow a one-size-fits-all approach, providing a fixed order of learning materials and expecting all learners to follow this recommended path. Thus, they neither motivate nor support their learners in adapting the courses to their individual preferences. In the work at hand, we tackle this issue by introducing and evaluating the concept of flexible learning paths in MOOCs. We, therefore, establish a network of dependencies between course content, omit intermediate deadlines, and thereby rethink the way learners interact with the course. By presenting learners with a non-linear course format, we encourage them to create their individual learning paths based on instructor-defined dependencies and their personal interests. Our evaluation of flexible learning paths within a programming MOOC shows that learners chose many different learning paths. Despite achieving similar results in individual tasks compared to learners using the traditional course structure, they engaged with less course content, resulting in a slight decrease in their overall performance. This may indicate a lack of self-regulatory learning skills, with learners struggling to organise their work without instructor-given deadlines. However, the flexible course format significantly increased the motivation of learners. By introducing and evaluating the concept of flexible learning paths in MOOCs, this work provides valuable insights into the individualisation of online education.

Despite advances in image recognition, recognizing novel categories in unlabeled data remains challenging for machine learning methods, even though humans can perform this task with ease. A recently developed setting to tackle this problem is Generalized Category Discovery (GCD), in which the task is to, given a labeled dataset, classify an unlabeled dataset, where the unlabeled dataset contains both known classes and novel classes that do not appear in the labeled data. Existing GCD methods mostly focus on learning strong image representations, on which they then apply a clustering algorithm such as k-means. Despite obtaining good performance, they do not fully exploit the potential of the learned features due to the simple nature of the clustering mechanism. To address this issue, we make use of the fact that local neighborhoods in self-supervised feature spaces are highly homogeneous. We leverage this observation to develop Guided Cluster Aggregation (GCA), a hierarchical approach that first groups the data into small clusters of high purity, then aggregates them into larger clusters. Experiments show that GCA outperforms semi-supervised k-means in most cases, especially in fine-grained classification tasks. Code available at https://github.com/J-L-O/guidedcluster-aggregation.

Fine-tuning and inference on Large Language Models like BERT have become increasingly expensive regarding memory cost and computation resources. The recently proposed computation-flexible BERT models facilitate their deployment in varied computational environments. Training such flexible BERT models involves jointly optimizing multiple BERT subnets, which will unavoidably interfere with one another. Besides, the performance of large subnets is limited by the performance gap between the smallest subnet and the supernet, despite efforts to enhance the smaller subnets. In this regard, we propose layer-wise Neural grafting to boost BERT subnets, especially the larger ones. The proposed method improves the average performance of the subnets on six GLUE tasks and boosts the supernets on all GLUE tasks and the SQuAD data set. Based on the boosted subnets, we further build an inference framework enabling practical width- and depth-dynamic inference regarding different inputs by combining width-dynamic gating modules and early exit off-ramps in the depth dimension. Experimental results show that the proposed framework achieves a better dynamic inference range than other methods in terms of trading off performance and computational complexity on four GLUE tasks and SQuAD. In particular, our best-tradeoff inference result outperforms other fixed-size models with similar amount of computations. Compared to BERT-Base, the proposed inference framework yields a 1.3-point improvement in the average GLUE score and a 2.2-point increase in the F1 score on SQuAD, while reducing computations by around 45%.

RCTs are an essential part of scientific research, allowing testing of different experimental setups with independent groups. In an eLearning context, this can relate to both platform features and teaching didactic experiments. So far, the MOOC platform we operate provided no simple way to run A/B tests at the course content level, but only at the platform feature level. However, this limitation is unacceptable when studying the effect of different content presentation styles and methods, so we had to develop a new RCT feature. In this paper, we describe our requirements and the challenges associated with content A/B testing in MOOCs. We examine how other platform providers solved these problems and compare our solution design. We then describe the technical implementation of the new feature, which difficulties we encountered on our complex infrastructure, and how we solved them. We also share our results from the first successful content A/B test experiments and further planned enhancements. The paper thus offers researchers the opportunity to better understand the technical concept and our implementation of content A/B testing on a MOOC platform. It helps developing a technical solution for RTCs on their infrastructure.

The digital transformation is challenging various areas of our everyday lives. Central aspects of our digital identities are our knowledge and experience, asserted by certificates, references and credentials. While digital credentials have primarily started to be used in the e-government and health sectors, acceptance is more and more transferring to the education sector. Within the area of digital credentials, a multitude of different projects and initiatives exist which are hard to follow and compare. Some researchers have attempted to perform systematic literature reviews. However, the scope of the review was often limited. Therefore, we create a conceptual framework to model the requirements of a digital education credential system by following both a conceptual-to-empirical approach and an empirical-to-conceptual approach. We perform an extensive literature review, focusing on identifying needs articulated in the relevant literature. We demonstrate the applicability and usability of the framework by developing a heatmap of requirements representing the current literature landscape. In a second step, we utilised the framework to interview subject matter experts and let them prioritise requirements for a digital education credential system. By comparing the heatmap results with the prioritisation of experts, we identified requirements, which are rarely found in the literature but rated as necessary by experts—making it most likely that they are typically overlooked when a new system is designed. We identified necessary, desirable and overlooked requirements and gave an indicator for prioritisation in our framework.

Pre-trained Language Models in Natural Language Processing have become increasingly computationally expensive and memory demanding. The recently proposed computation-adaptive BERT models facilitate their deployment in practical applications. Training such a BERT model involves jointly optimizing subnets of varying sizes, which is not easy due to their mutual interference with one another. The larger-size subnets in particular could deteriorate when there is a large performance gap between the smallest subnet and the super-net. In this work, we propose Neural grafting to boost BERT subnets, especially the larger ones. Specifically, we regard the less important sub-modules of a BERT model as less active and reactivate them via layer-wise Neural grafting. Experimental results show that the proposed method improves the average performance of BERT subnets on six datasets of GLUE benchmark. The subnet performing comparable to the supernet BERT-Base reduces around 67% and 70% inference latency on GPU and CPU, respectively. Moreover, we compare two Neural grafting strategies under varied experimental settings, hoping to shed light on the application scenarios of Neural grafting.

Biometric recognition systems are an integral part of security in modern computer systems. Contrary to related work, our approach to biometrics does not base on individual patterns in sensor data. We measure visual self-recognition as an in-brain identity validation mechanism and aim to employ it as a biometric trait. This work builds on a public data set of an eye-tracking study investigating the visual self-recognition of 116 volunteers. Exploration of this high-quality data revealed significant effects of self-recognition in the time course of pupil size and microsaccade generation. These findings facilitated the extraction of features to train four different experimental classification models. The best-performing model is a pre-trained classifier on integrated data by achieving an accuracy of 0.82. Our research supports the hypothesis that self-recognition can be a feasible biometric trait. Furthermore, we found that self-recognition can be measured using eye-tracking devices. Future work on model optimization and additional feature extraction is required. It will enable to introduce technologies utilizing self-recognition to real-world applications.

If a researcher wants to contribute in the field of biometric authentication systems (BAS), much manual work is required to gather the current state-of-research (SoR) for a specific type of BAS. Due to loosely structured scientific papers as result reports, all relevant details need to be extracted, aggregated and collected individually. Once the SoR is determined, an exchange with other researchers is hardly possible due to a missing exchange format and they usually have to start from scratch repeating the same inefficient procedure often targeting the same papers. This work presents a domain model for biometric authentication systems and their evaluations, BASE, as a solution. It models and interconnects biometric characteristics, the target architecture, evaluation scenarios, and evaluation criteria grants to represent the most common parts of a BAS. For evaluation, we map 12 BAS approaches from related work into BASE and determine mainly low mapping efforts. In this context, we first identified diverse subsets of BASE being inferable from the analyzed papers. Secondly, most researchers still target rather broad evaluation categories like security or usability instead specific criteria within these categories where the latter would simplify the basic assessment of a BAS. Overall, we consider BASE a promising step to improve the efficiency in SoR determination.

Pre-trained Language Models bring about an in-creasing computational and memory cost. The recently proposed computation-flexible BERT models facilitate their deployment in varied computational environments. Training such flexible BERT models involves jointly optimizing multiple BERT subnets that inevitably interfere with one another. Besides, the performance of large sub nets is limited when there is a significant performance gap between the smallest sub net and the supernet, despite methods managing to enhance the smaller subnets. We propose layer-wise Neural grafting to boost BERT subnets, particularly the larger ones. The proposed method improves the average performance of BERT sub nets on six out of eight GLUE tasks. Furthermore, we build a flexible BERT model that enables practical width- and depth-dynamic inference regarding different inputs by combining width-dynamic gating modules and early exit off-ramps in the depth dimension. Experimental results demonstrate that the proposed framework achieves a better dynamic inference range than other methods in the trade-off between performance and computational complexity on four GLUE tasks and the SQuAD data set. Our optimal-tradeoff inference result, in particular, outperforms related fixed-size models with comparable computational complexity. Compared to the supernet, BERT-Base, this inference result improves the average GLUE score and Fl score on SQuAD by 1.3 and 2.2 absolute points, respectively, and decreases computations by around 45%.

Mutual knowledge distillation (MKD) is a technique used to transfer knowledge between multiple models in a collaborative manner. However, it is important to note that not all knowledge is accurate or reliable, particularly under challenging conditions such as label noise, which can lead to models that memorize undesired information. This problem can be addressed by improving the reliability of the knowledge source, as well as selectively selecting reliable knowledge for distillation. While making a model more reliable is a widely studied topic, selective MKD has received less attention. To address this, we propose a new framework called selective mutual knowledge distillation (SMKD). The key component of SMKD is a generic knowledge selection formulation, which allows for either static or progressive selection thresholds. Additionally, SMKD covers two special cases: using no knowledge and using all knowledge, resulting in a unified MKD framework. We present extensive experimental results to demonstrate the effectiveness of SMKD and justify its design.

The popularity of Massive Open Online Courses (MOOCs) as a means of delivering education to large numbers of students has been growing steadily over the last decade. As technology improves, more educational content is becoming readily available to the public. JupyterLab, an open-source web-based interactive development environment (IDE), is also becoming increasingly popular in education, however, it is so far primarily used in small classroom settings. JupyterLab can provide a more interactive, hands-on, and collaborative learning experience for students in MOOCs, and it is highly customizable and can be accessed from anywhere. To capitalize on these benefits, we have developed OpenJupyter, which integrates JupyterLab at scale with MOOCs, enhancing the student learning experience and providing hands-on exercises for data science courses, making them more interactive and engaging. While MOOCs provide access to education for a large number of students, one of the significant challenges is providing effective and timely feedback to learners. OpenJupyter includes an auto-assessment capability that addresses this problem in MOOCs by automating the evaluation process and providing feedback to learners in a timely manner. In this paper, we provide an overview of the architecture of OpenJupyter, its scalability in the context of MOOCs, and its effectiveness in addressing the auto-assessment challenge. We also discuss the Advantages and limitations associated with using OpenJupyter in a MOOC context and provide a reference for educators and researchers who wish to implement similar tools. Our efforts aim to foster an open educational environment in the field of programming by providing learners with an interactive learning tool and a streamlined technical setup, allowing them to acquire and test their knowledge at their own pace.

Podcasts are a widely-used medium for communication and learning. One advantage of them is the possibility to pursue other activities while listening. Contrasting, Massive Open Online Courses (MOOCs) employ video-based teaching methods. Current research, however, challenges the interactivity and variation of teaching content in established MOOCs. This manuscript presents an experiment conducted with a podcast series deployed alongside a MOOC on cybersecurity. In our Static-Group Comparison, we identified a significant increase in learning success in weekly graded exercises (6.3%) and the course's final examination (6.4%) for learners exposing themselves to the podcast. Our first study results are promising in favor of multimedia learning. Hence, we present ideas for additional analysis and briefly outline which aspects of the results should be discussed in more depth.

Identifying persons based on their interaction with a door handle has been an active research field in the past years and related work presented promising results. In this context, they instrumented door handles with sensors like accelerometer, gyroscope, or touch and extracted features from complete opening interactions that include pushing down the handle, opening the door, entering the room, and releasing the handle. This work extends existing research with an evaluation whether already the initial door handle’s push-down-phase (PDP) provides sufficient data for behavior-based authentication. For that, recorDOOR was built, a door handle prototype equipped with a camera, accelerometer, gyroscope, four touch stripes, and three ultrasonic distance sensors. We recorded 768 door interactions from eight participants (48 per hand) using recorDOOR, manually labeled the PDP’s start and end, and when the door started moving using our camera stream. We are the first that recorded both hands and also simulated four real-life door opening scenarios including a novel opening careful scenario when standing in front of a door. Our quantitative analysis of the extracted PDPs shows that there is a high variance between PDP durations from less than 40 milliseconds to more than a second and that there are notable differences between left and right hand while opening independent of the scenario. Furthermore, over 95% of our participants started moving the door before the handle was pushed down to its maximum that further shortens the time for a user identification. Based on 235 extracted features and nested 5-fold cross-validation, a Gradient Boosting Classifier performed best and achieved a mean accuracy of 87.92% in an identification scenario for the overall PDP and 82.41% if only considering the PDP before the door is already moved.

Anomaly detection is a significant problem that has been researched within diverse research areas and application domains, especially in the area of web-based internet services or cybersecurity. Many anomaly detection techniques have been developed for specific application domains, while others are more generic. The Log files of Web-server give insight into the state of web-server and applications running on it and enable the detection of abnormal incidents or behavior. This paper focuses on particularly web-server HTTP logs to the problems of Web-server Log Anomaly Detection (WLAD) due to their own nature and features and aims to provide a brief review of different Data-driven techniques to get to the bottom of recent studies and developments made in the context of WLAD. Moreover, in this paper, the literature related to webserver logs analysis, as well as other closely related to the WLAD topic, are taken into consideration for review. We have classified existing techniques into different categories based on the underlying approach adopted. When applying a particular technique, these assumptions can be used as guidelines to assess the method’s effectiveness in this area. We also provide a basic security anomaly detection approach for each category and compare the existing methods as variants of the basic technique. Further, we identify the cons and pros of the current practices for each category. We also discuss the computational complexity of the methods, which is an essential issue in the domain of Big Data.

Massive Open Online Courses (MOOCs) have become a well-established tool for lifelong learning since their first invention in 2008. Although the courses initially aimed at high social interaction rates and interactive course structures, most conducted courses since the big MOOC hype in 2012 follow a classic teaching approach that often lacks interactive elements and suffers from low user engagement. To make our course format more engaging and interactive, we experimented with gameful learning by creating two storified Java programming MOOCs, which include a detective story that spans all course weeks. This paper evaluates learner engagement and perception with story videos and interactive story quizzes in our storified programming courses. We found that the learners’ engagement with the optional story videos and quizzes is slightly less than regular video lectures and other mandatory learning materials. Nevertheless, some learners are particularly motivated by the story and interact with the story elements throughout the course. These learners have relatively good learning outcomes; for some, the story is even their primary motivation to continue with the course. However, some learners dislike the story elements and feel distracted by them. In terms of acceptance, we did not notice any major differences in demographic factors like age and gender. However, we found that learners motivated to participate in the course above average and even using additional bonus tasks show no higher interest in the story than the average learner. Thus, we disprove that learners only use the story because they use all offered course elements but because they are actually interested in it.

Massive Open Online Courses (MOOCs) are usually attended by several thousand learners who barely get to know each other during the course period. Being unaware of fellow learners often results in a low sense of community. In addition, many MOOC learners are afraid of using the course forum, which often is the only participation opportunity in social course activities apart from forming smaller learning groups. Thus, learners can easily be frustrated with the course content when feeling alone. To improve social presence and the sense of community, course instructors can use ice-breaking games. First, this paper evaluates which kind of ice-breaking games can be used in MOOCs. Afterward, we present the results from a first experiment where we use “self-reflection sociograms as an icebreaking activity. Most learners perceived the implemented Self-Reflection Questionnaires” (SRQ) ice-breaker as a positive course feature (68.35%). SRQs increased the sense of community, and learners were satisfied (91.06%) with their perceived community sense level. The SRQs were also helpful for the teaching teams. Our results indicate that further investigation of SRQs is beneficial to explore the provided value for course instructors and their influence on individual MOOC learners and community-building.

One of the most pressing problems in the automated analysis of historical documents is the availability of annotated training data. The problem is that labeling samples is a time-consuming task because it requires human expertise and thus, cannot be automated well. In this work, we propose a novel method to construct synthetic labeled datasets for historical documents where no annotations are available. We train a StyleGAN model to synthesize document images that capture the core features of the original documents. While originally, the StyleGAN architecture was not intended to produce labels, it indirectly learns the underlying semantics to generate realistic images. Using our approach, we can extract the semantic information from the intermediate feature maps and use it to generate ground truth labels. To investigate if our synthetic dataset can be used to segment the text in historical documents, we use it to train multiple supervised segmentation models and evaluate their performance. We also train these models on another dataset created by a state-of-the-art synthesis approach to show that the models trained on our dataset achieve better results while requiring even less human annotation effort.

The Hasso Plattner Institute (HPI) has been successfully delivering courses on several MOOC (Massive Open Online Course) platforms for the last 10 years, offering courses on various topics in the context of Artificial Intelligence (AI), Machine Learning (ML), and Data Science. In recent years, Jupyter Notebooks have become one of the most widely used tools for data science applications, a platform for learning and practicing various programming languages. We want to integrate JupyterHub into our learning platform in order to provide students with hands-on experience in AI. We have conducted a survey with a series of research questions in order to understand the needs of instructors in their courses at different institutions. In this paper, we present a detailed analysis of our survey results and we discuss our future approach to using JupyterHub as an infrastructure to solve hands-on programming exercises on our platform. We propose the idea of creating a tool to automate server and environment creation for students to work on. This tool would give instructors a platform to operate from and allow them to customize their courses. Moreover, it would help them automate assignment submissions, grading, and provide feedback to their students.

In 2017, Geoffrey West published his book "Scale" in which he examined universal laws of scale in different contexts. Inspired by his keynote in 2021's [email protected] conference, we investigated the applicability of these laws in the context of Massive Open Online Courses and learners' behavior. We tested these laws on different learning platforms from academic, enterprise and social, and research contexts. In this paper, we examine course characteristics, such as course size, the completion rate, and the forum activity. We observed that the number of issued certificates scales almost identically on all examined platforms, while forum participation scales slightly different on each of the platforms. In the future, we will perform a deeper analysis on the forum behavior that exceeds a mere quantitative analysis.

Whether a lecturer presence helps or hinders in audio-visual learning material has been raised numerous times. While previous research found no substantial evidence in favor of a lecturer presence in controlled eye-tracking experiments, the given study analyzes results of a field-experiment in a German MOOC with 2,938 active participants taking a four-week course on data structures and algorithms. The research team produced specific content for this experiment with the goal to compare traditional slide-lectures with a modern explainer video style as seen on major video platforms. These two treatment groups are identical on the audio track and truly only differ in terms of the visual experience of the lecture on recursion. The variables are: 1) Perception of a learner defined by content, the speaker and his/her own learning and 2) Scores of a recall and transfer assessment. The first variable is conducted by a user survey (n=490), the skill assessment is measured by two posttests (quiz & programming task). The findings indicate a different perception of the speaker's focus, a significant better evaluation of the lecturer condition and higher scores on the recall posttest. No difference is seen in the perceived degree of professionalism, the self-reported level of attention and the scores of the transfer task. By testing and replicating previous findings in a real MOOC setting with adult learners, the given study contributes to the research of video-based learning in general, and to the sub-topic of effective teaching settings for computer science concepts in scaling environments.

Abstract Live migration is an essential feature in virtual infrastructure and cloud computing datacenters. Using live migration, virtual machines can be online migrated from a physical machine to another with negligible service interruption. Load balance, power saving, dynamic resource allocation, and high availability algorithms in virtual data-centers and cloud computing environments are dependent on live migration. Live migration process has six phases that result in live migration cost. Several papers analyze and model live migration costs for different hypervisors, different kinds of workloads and different models of analysis. In addition, there are also many other papers that provide prediction techniques for live migration costs. It is a challenge for the reader to organize, classify, and compare live migration overhead research papers due to the broad focus of the papers in this domain. In this survey paper, we classify, analyze, and compare different papers that cover pre-copy live migration cost analysis and prediction from different angels to show the contributions and the drawbacks of each study. Papers classification helps the readers to get different studies details about a specific live migration cost parameter. The classification of the paper considers the papers’ research focus, methodology, the hypervisors, and the cost parameters. Papers analysis helps the readers to know which model can be used for which hypervisor and to know the techniques used for live migration cost analysis and prediction. Papers comparison shows the contributions, drawbacks, and the modeling differences by each paper in a table format that simplifies the comparison. Virtualized Data-center and cloud computing clusters admins can also make use of this paper to know which live migration cost prediction model can fit for their environments.

The world is experiencing remote interaction in unprecedented frequency, as people stay in touch and work together remotely in pandemic times. IT solutions for remote work such as videoconferencing systems have received a lot of critical attention as they seem to induce fatigue. By contrast, in online video games, people collaborate passionately and energetically for hours at a time. In this paper, we introduce methodological frameworks for studying the impact of online video games on individual thinking capacities and team collaboration, with the goal of inspiring IT solutions for remote work. The research is grounded in neurodesign, an approach that uses neuroscientific research to underpin the analysis of how digital technology impacts humans. A major focus of this paper is body motion and virtual environments, tracing how they impact processes of team formation and creative thinking capacities, both during play and shortly thereafter. The paper reports on three pilot studies and methodological developments. The findings indicate that remote interaction leads to increased team cohesion and creative team performance, in particular when the interaction involves synchronous motion with a team partner, such as driving next to each other in Mario Kart on Nintendo Switch.

Lockdowns, short-time work, home office, school closures, and contact restrictions caused by the Corona pandemic have changed work, life, and education worldwide. Since it is still mainly women who provide childcare, a discussion about re-traditionalization has emerged in the context of the Corona crisis. Up to now, there are hardly any empirical findings on gender effects of the Corona pandemic on participation in further education. We make a contribution to this political, societal as well as economical relevant topic by analyzing participation, learning efforts, and completion rates in Massive Open Online Courses (MOOCs) during the Corona pandemic with a focus on gender aspects. Based on micro theoretical approaches as well as hypotheses of social and economic influences on participation in continuing education, we quantitatively examine the access as well as participants' engagement and success to more than ten MOOCs in Science, Technology, Engineering, and Mathematics (STEM) offered during the Corona-related restrictions in 2020 and one year before by gender. Based on descriptive statistics, we register increasing course enrollments, but a decreasing share of female learners since the Corona-related restrictions. Nevertheless, we Figure out that both participating men as well as women access more digital learning materials and reach higher completion rates than before the Corona crisis. Our multivariate linear probability models indicate further influences of socio-economic variables, such as career status. With regard to the current state of research, we discuss our findings, draw conclusions, formulate implications, and give an outlook on our future research.

Social media has become a popular communication platform on which shared content such as images form a large part of the communicated data. Yet, shared images can reveal sensitive information in the sense that the data after its publication remains accessible. Existing studies provide mechanisms to modify co-owned images for user privacy but require that every user involved be online in order to reach an agreement. In cases where users are offline at the time when the image is posted, no privacy agreement can be reached. Having a method of reaching a privacy agreement even when some of the users in the co-owned image are offline is useful in enforcing individual privacy settings vis-a-vis the co-owned image. In this paper, we present a multi-agent negotiation model that enforces individual privacy settings with respect to co-owned images even when the users are offline. Our multi-agent model includes three components, namely a coordinator agent, predictor agent, and filtering algorithm. The coordinator agent collects users’ opinions vis-a-vis a co-owned image to form an image that expresses the opinions of the involved users. The predictor agent supports the expression of offline user opinions, while the filtering algorithm removes privacy-violating information with respect to recent user opinions. Results from our proof-of-concept implementation indicate that improved efficiency in terms of privacy decisions can be achieved by employing agents to support offline user decisions regarding shared content.

In this research, we formulate malware detection as a large-scale data-mining problem within Security Information and Event Management (SIEM) systems. We hypothesize that behavioral analysis of executable/process activities, such as file reads/writes, process creations, network connections, or registry modifications, enables the detection of advanced stealthy malware. To achieve this detection, we model processes behaviors as a set of directed acyclic graph streams and identify outliers in the set of graph streams. We enable this detection by conversion of the behavioral graph streams into documents, embedding using state-of-the-art Natural Language Processing model, and eventually performing novel outlier detection on the high dimensional vector representation of the documents. We evaluate our approach in a real-world setting, next to the SIEM system of a large-scale international enterprise (over 3TB of EDR logs). The proposed method has shown the capability to detect previously unknown threats.

Identifying people while entering a room is required to control access to the room but also for personalization in the context of the Internet of Things. Over the last years, several research proposed ways for identification based on the individual door interaction using touch, motion, or computer vision data. Although they reported promising identification accuracies, the results were based on supervised artificial door interaction scenarios were many real world situations are not covered and a decent recording setup is missing yet. This work presents and evaluates DoorCollect, a system to collect interaction data at door handles in an unsupervised way with the help of automated interaction detection and tagging based on BLE beacons, and an alerting and monitoring functionality. Based on a 22 day recording session, we underline its capabilities but also identify further improvement options such as an additional camera-based data verification or a more sensible alerting.

Students can use personal mobile devices to access Massive Open Online Courses (MOOCs) in addition to desktop computers. However, user interfaces are often only scaled to smaller screen sizes and interaction patterns of a desktop learning experience do not always fit well with the characteristics of mobile devices. Adequate solutions for answering self-test questions on mobile devices often do not exist. In this paper, we explore the currently shown interaction patterns of MOOC learners when answering self-tests to make an informed decision about the requirements for the appropriate solution on mobile devices. The students' context was categorized into Desktop Web, Mobile Web, and Mobile Application. In an observational case study, the interaction events of two courses were analyzed regarding these device groups. Desktop Web is the most used environment. No practical differences between device groups were identified for subsequent attempts. Learners stick to a single device group and often only participate once in a self-test. Also, learners using mobile applications spend more time submitting self-tests.

Learners in Massive Open Online Courses offering practical programming exercises face additional challenges next to the actual course content. Beginners have to find approaches to deal with misconceptions and often struggle with the correct syntax while solving the exercises. The paper at hand presents insights from offering contextual tips in a web-based development environment used for practical programming exercises. We measured the effects of our approach in a Python course with 6,000 active students in a hidden A/B test and additionally used qualitative surveys. While a majority of learners valued the assistance, we were unable to show a direct impact on completion rates or average scores. We however noticed that users requesting tips took significantly longer and made more use of other assistance features of the platform than users in our control group. Insights from our study can be used to target beginners with more specific hints and provide additional, context-specific clues as part of the learning material.

Streaming games and entertainment content are established formats on large portals like YouTube and Twitch. Educational streams have not yet reached the same popularity. Consequently, the existing research of gaming streams by far exceeds the research of live-streaming lectures. In this paper, we contribute first insights regarding this area, outline the status quo of the edu-streaming ecosystem, and highlight common approaches and characteristics. Through a descriptive study of 100 popular gaming streams, we systemize the inductively found features and approaches that are seen in both ecosystems. With a further focus on 20 educational streams, we highlight features such as synchronous community building and on-stream interactivity. We project the main differences between the core characteristics of MOOCs and educational streams. Finally, we propose further research directions for the emerging field of public, synchronous online education.

The current Self-Sovereign Identity (SSI) ecosystem is rapidly changing and ill-defined. Manifold actors, projects, and initiatives produce different SSI solutions, frameworks, protocols, and distributed ledgers. Even though some patterns exist among SSI ecosystem members, no elaborate systematization has been made. This paper conducts a systematic gray literature review to structure the SSI ecosystem. Specifically, we derive a four-dimensional taxonomy that portrays members of the SSI ecosystem. Then, we classify the ecosystem members into eight archetypes. The goals are to allow researchers to describe SSI ecosystem members, help new and existing members locate themselves within the SSI ecosystem, and provide an overview of members’ functionalities. We find that SSI ecosystem members either govern the SSI ecosystem and/or networks, implement SSI offerings, or support governing and/or implementing members. The study suggests that, as the SSI ecosystem grows, the number of governing members will grow slower than the number of implementing and supporting members.

The Self-Sovereign Identity (SSI) paradigm postulates global unique identities that are controlled by the user. To achieve a widespread applicability, the emphasized interoperability principle supports the proclaimed ambition. Furthermore, identity portability enables the transfer of the identity to another SSI solution. These axioms gain additional momentum due to the development of numerous implementations. In this paper, we examine interoperability and portability concepts for SSI. Initially, we define these principles regarding the blockchain-based SSI model. Subsequently, we outline assessment criteria considering functional scope, governance/ trust, scalability and further characteristics. For interoperability, we evaluate the concepts of protocol and standard, broker, hub and pairing. Besides that, we assess the transformer and auxiliary solutions for portability. We can conclude that all interoperability schemes provide the maximum functional level theoretically. In contrast, portability patterns are fragmented in this regard. Nonetheless, protocol and standards can only be applied in the design phase, whereas broker, hub, pairing, transformer and auxiliary solutions enable interoperability, respectively portability post-deployment of the SSI system.

Binary Neural Networks (BNNs) are neural networks which use binary weights and activations instead of the typical 32-bit floating point values. They have reduced model sizes and allow for efficient inference on mobile or embedded devices with limited power and computational resources. However, the binarization of weights and activations leads to feature maps of lower quality and lower capacity and thus a drop in accuracy compared to their 32-bit counterparts. Previous work has increased the number of channels or used multiple binary bases to alleviate these problems. In this paper, we instead present an architectural approach: MeliusNet. It consists of alternating a DenseBlock, which increases the feature capacity, and our proposed ImprovementBlock, which increases the feature quality. Experiments on the ImageNet dataset demonstrate the superior performance of our MeliusNet over a variety of popular binary architectures with regards to both computation savings and accuracy. Furthermore, BNN models trained with our method can match the accuracy of the popular compact network MobileNet-v1 in terms of model size and number of operations. Our code is published online: https://github.com/hpi-xnor/BMXNet-v2.

Objective and interpretable metrics to evaluate current artificial intelligent systems are of great importance, not only to analyze the current state of such systems but also to objectively measure progress in the future. We propose a novel metric, called Fuzzy Topology Impact (FTI), that assesses both the quality and diversity of a generated set using topological representations combined with fuzzy logic. In our synthetic experiments, FTI consistently outperforms current evaluation methods in terms of stability and sensitivity to detect drops in quality and diversity in the generated set, both on image and text generation tasks. Moreover, FTI shows a high degree of correlation to human evaluation on unconditional language generation.

Creativity is a central phenomenon in human life. World-famous scientists and artists are praised for their creative genius. Schools and universities seek to educate creativity in students and many employers want to hire creative personnel. However, the measurement of creativity is difficult up to the present day. Standard creativity tests typically require human expertise in the evaluation of test responses. This evaluation is often more time-intensive than taking the test itself. Moreover, creativity tests are still regularly conducted in a pen-and-paper format, rendering the data analysis all the more tedious. In this article, we propose a digital game for assessing creativity. It can be hosted online. The data analysis can be automated and conducted in real-time. The test is implemented as a tower defense game ("Immune Defense"). We submit that a video game constitutes a natural setting, as opposed to a formal testing scenario, and provides an opportunity to test real-life creativity. We use the game event data gathered during each round of the game to determine a player's creativity score. A study with 17 participants was performed to compare game-based creativity scores to scores obtained with a standard creativity test, the Alternative Uses Task. Our preliminary data validate the proposed approach.

We propose a family of metrics to assess language generation derived from population estimation methods widely used in ecology.More specifically, we use mark-recapture and maximumlikelihood methods that have been applied over the past several decades to estimate the size of closed populations in the wild.We propose three novel metrics: ME Petersen and ME CAPTURE , which retrieve a single-valued assessment, and ME Schnabel which returns a double-valued metric to assess the evaluation set in terms of quality and diversity, separately.In synthetic experiments, our family of methods is sensitive to drops in quality and diversity.Moreover, our methods show a higher correlation to human evaluation than existing metrics on several challenging tasks, namely unconditional language generation, machine translation, and text summarization.

Learners in Massive Open Online Courses (MOOCs) are reiterating over the provided course material especially self-tests - to consolidate their knowledge. This is a manual and often cumbersome process as MOOC platforms do not provide personalized revision opportunities. This paper introduces the design and concept of a flashcard-like recap tool based on spaced repetition learning techniques. The recap material is derived from existing self-test questions. The usage rates of the recap tool were observed in three courses and peaked before graded assignments, primarily before the final exam. When choosing the question quantity, learners preferred either the smallest option or wanted to revise all of the available questions, whereas the average number of questions per recap session increases over time. Recap tool users who completed a recap session showed smaller error rates than those who stopped a recap session abruptly, while learners who skipped questions performed worst. Course participants who used the recap tool throughout the course achieved on average more of the available points. Statistically highly significant differences were detected for all observed courses. An additional survey (N=79) gathered qualitative feedback and impressions from the learning community.

Recent sociotechnical innovations in online communication make user-content interaction, dynamic user-user interaction and user-generated content interesting for content analysis, but raise complex challenges of reliable and valid coding as well as content analytical interpretations. We approach these new issues by analyzing the asynchronous communication and discourses in an exemplary Massive Open Online Course (MOOC) in IT security with 4,203 enrollments and 1,343 posts on 192 topics in the forum by 274 different users. We analyze the user-driven contributions to forum discussions using mixed methods and answer the question of whether men and women raise different questions in the discussion forums and if yes, with what kind of effects. We show that many women prefer closed and consent questions, as well as rhetorical questions over discussion, hypothetical questions, and contact requests. The question type significantly influences the way of answering: Direct and open questions particularly encourage fellow learners to answer in a helpful way and in a short time of two to three hours. Responses to questions asked by women were characterized more often by detailed explanations, a simple reply or expression of appreciation, and less often with vagueness. In other words, no one repeated the initial questions asked by women and none of their questions were met with a reaction of amazement. The qualitative part of our content analysis revealed that women’s discussion posts imply more often uncertainty or a slightly lower level of self-efficacy than men’s contributions. Forum discussions took place particularly in the afternoon from 2 to 6 p.m. CET, a time period when many women have other obligations, therefore making it less than ideal. We conclude by making recommendations based on our findings for improvements in supporting underrepresented groups in online learning in science, technology, engineering, and mathematics (STEM) and discuss potential limitations of our case study.

Peer assessment has become a regular feature of many MOOC1platforms and also has potential for other contexts where learning and teaching are required to scale because of growing numbers of students. Where manual grading is not possible due to the large number of submissions and the tasks to be assessed are to complex or open-ended to be assessed by machines, peer assessment offers a valuable alternative. However, particularly in the context of MOOCS, courses featuring peer assessments often have lower completion rates. Furthermore, participants with negative expectations and opinions about this form of assessment are generally ‘louder’ in their communication with the teaching teams than their counterparts who respond more positively. We have, therefore, set out to establish a broader understanding how the participants perceive the appropriateness and effectiveness of peer assessed tasks, and the quality of the received reviews on the X1, X2, and X32MOOC platforms. For this purpose, we have conducted post-course surveys in a large number of courses that included peer assessments. Additionally, we analyzed the discussions in the forums of these courses as the post-course surveys often are biased due to the low proportion of unsuccessful participants that are still around at the end of a course.1Massive Open Online Course2The platforms’ names have been obscured for double-blind review

Natural Language Generation (NLG) task is to generate natural language given structured data, like Meaning Representations (MRs). The generated sentences are supposed to convey all the information in MRs and be fluent and realistic as human written. The constantly emerging pre-trained models push the state-of-the-art performance of various Natural Language Processing tasks to a new level. The common practice of fine-tuning pre-trained language models for NLG is to regard it as a text-to-text generation task. That is, MRs are converted into graph structures, then the graphs are linearized as text sequences, during which lots of pre-processing and post-processing work are required. We explore different methods to organize the MRs and show that just linearizing the information in MRs achieve decent results, while complex annotation process can be omitted. Under the circumstances, further experiments demonstrate that the fine-tuned pre-trained model achieves comparable results with state-of-the-art models on the RNNLG dataset.

Nowadays processing of Big Security Data, such as log messages, is commonly used for intrusion detection purposes. Its heterogeneous nature, as well as a combination of numerical and categorical attributes, do not allow to apply the existing data mining methods directly on the data without feature preprocessing. Therefore, a rather computationally expensive conversion of categorical attributes into vector space should be utilised for analysis of such data. However, a well-known k-modes algorithm allows to cluster the categorical data directly and avoid conversion into the vector space. The existing implementations of k-modes for Big Data processing are ensemble-based and utilise two-step clustering, where data subsets are first clustered independently, whereas the resulting cluster modes are clustered again in order to calculate metamodes (or "modes of modes") valid for all data subsets. In this paper, the novel frequency-based distance function is proposed for the second step of ensemble-based k-modes clustering. Besides this, the existing feature discretisation method from the previous work is utilised in order to adapt k-modes for processing of mixed data sets. The resulting k-metamodes algorithm was tested on two public security data sets and reached higher effectiveness in comparison with the previous work.

Massive Open Online Courses (MOOCs) provide the opportunity to offer free and open education at scale. Thousands of students with different social and cultural backgrounds from all over the world can enroll for a course. This diverse audience comes with varying motivations and intentions from their personal or professional life. However, course instructors cannot offer individual support and guidance at this scale and therefore usually provide a one-size-fits-all approach. Students have to follow weekly-structured courses and their success is measured with the achievement of a certificate at the end. To better address the varying learning needs, technical support for goal-oriented and self-regulated learning is desired but very limited to date. Both learning strategies are proven to be key factors for students' achievement in large-scale online learning environments. Therefore, this paper presents a continuative study of personalized learning objectives in MOOCs to encourage goal-oriented and self-regulated learning. Based on the previously well-perceived acceptance and usefulness of the concept of personalized learning objectives, this study examines which learners select an objective and how successful they complete objectives. Concerning the learners' socio-demographic and geographical background, we could not identify any practical significant difference between students with selected learning objectives and the total course population. However, we have identified promising objective achievement rates, and we have observed a practical significant improvement of the certification rates comparing the total course population and students who selected an objective that included a graded certificate. This has also demonstrated a method for calculating more reasonable completion rates in MOOCs.

Teamwork and graded team assignments in MOOCs are still largely under-researched. Nevertheless, the topic is enormously important as the ability to work and solve problems in teams is becoming increasingly common in modern work environments. The paper at hand discusses the reliability of a system to detect free-riders in peer assessed team tasks.

Language models trained with Maximum Likelihood Estimation (MLE) have been considered as a mainstream solution in Natural Language Generation (NLG) for years. Recently, various approaches with Generative Adversarial Nets (GANs) have also been proposed. While offering exciting new prospects, GANs in NLG by far are nevertheless reportedly suffering from training instability and mode collapse, and therefore outperformed by conventional MLE models. In this work, we propose techniques for improving GANs in NLG, namely Best Student Forcing (BSF), a novel yet simple adversarial training mechanism in which generated sequences of high quality are selected as temporary ground-truth to further train the generator. We also use an ensemble of discriminators to increase training stability and sample diversity. Evaluation shows that the combination of BSF and multiple discriminators consistently performs better than previous GAN approaches over various metrics, and outperforms a baseline MLE in terms of Fr ́ech ́et Distance, a recently proposed metric capturing both sample quality and diversity.

Binary and quantized neural networks are a promising technique to run convolutional neural networks on mobile or embedded devices. BMXNet 2 is an open-source framework that provides a broad basis for academia and industry. It provides a modern implementation of binary and quantized layers with a wide array of implemented state-of-the-art models. Our implementation fosters reproducibility of other works and our own work through publishing model code, hyperparameters, detailed model graphs, and training logs. Furthermore, we implement several applications for BNNs, including demo applications, which can run on a smartphone or a Raspberry Pi. The code can be found online: https://github.com/hpi-xnor/BMXNet-v2

For many years username and password are the common solution to protect sensitive web services despite its various drawbacks. While many alternatives were proposed to improve security, passwords are still included in any authentication procedure today. With the proposal of FIDO and FIDO2/WebAuthn a very promising approach was presented in the last years that may replace passwords at some time since it also enables password-less authentication in addition to work as a second factor for any web authentication. Although FIDO2/WebAuthn solves many problems of passwords using public/private key cryptography and the possibility to use strong authentication mechanisms like biometrics, it is still not capable of detecting an attacker once a successful login has happened. In this paper we evaluate the extension of FIDO2/WebAuthn to enable continuous authentication in the web. While this extension would enable the many proposals of continuous authentication for system or device protection to be used for web authentication, it allows the exchange of the relying parties' security requirements on the one hand and the authenticator's capabilities on the other hand, too. We evaluate our extension using an Android-based roaming authenticator communicating via Bluetooth Low Energy and show that the FIDO2/WebAuthn extension mechanism is suitable. While a real world deployment would require modifications in the different browser implementations, we further point out the challenges resulting from the different implementation levels and the high dynamics in the standard development such as different notification windows or parallelism issues.

We propose to tackle the mode collapse problem in generative adversarial networks (GANs) by using multiple discriminators and assigning a different portion of each minibatch, called microbatch, to each discriminator. We gradually change each discriminator's task from distinguishing between real and fake samples to discriminating samples coming from inside or outside its assigned microbatch by using a diversity parameter α. The generator is then forced to promote variety in each minibatch to make the micro-batch discrimination harder to achieve by each discriminator. Thus, all models in our framework benefit from having variety in the generated set to reduce their respective losses. We show evidence that our solution promotes sample diversity since early training stages on multiple datasets.

Social networking sites (SNS) are a rich source of latent information about individual characteristics. Crawling and analyzing this content provides a new approach for enterprises to personalize services and put forward product recommendations. In the past few years, commercial brands made a gradual appearance on social media platforms for advertisement, customers support and public relation purposes and by now it became a necessity throughout all branches. This online identity can be represented as a brand personality that reflects how a brand is perceived by its customers. We exploited recent research in text analysis and personality detection to build an automatic brand personality prediction model on top of the (Five-Factor Model) and (Linguistic Inquiry and Word Count) features extracted from publicly available benchmarks. Predictive evaluation on brands’ accounts reveals that Facebook platform provides a slight advantage over Twitter platform in offering more self-disclosure for users’ to express their emotions especially their demographic and psychological traits. Results also confirm the wider perspective that the same social media account carry a quite similar and comparable personality scores over different social media platforms. For evaluating our prediction results on actual brands’ accounts, we crawled the Facebook API and Twitter API respectively for 100k posts from the most valuable brands’ pages in the USA and we visualize exemplars of comparison results and present suggestions for future directions.

In this paper, we formulate threat detection in SIEM environments as a large-scale graph inference problem. We introduce a SIEM-based knowledge graph which models global associations among entities observed in proxy and DNS logs, enriched with related open source intelligence (OSINT) and cyber threat intelligence (CTI). Next, we propose MalRank, a graph-based inference algorithm designed to infer a node maliciousness score based on its associations to other entities presented in the knowledge graph, e.g., shared IP ranges or name servers.

Virtualization became a commonly used technology in datacenters during the last decade. Live migration is an essential feature in most of the clusters hypervisors. Live migration process has a cost that includes the migration time, downtime, IP network overhead, CPU overhead and power consumption. This migration cost cannot be ignored, however datacenter admins do live migration without expectations about the resultant cost. Several research papers have discussed this problem, however they could not provide a practical model that can be easily implemented for cost prediction in VMware environments. In this paper, we propose a machine learning approach for live migration cost prediction in VMware environments. The proposed approach is implemented as a VMware PowerCLI script that can be easily implemented and run in any vCenter Server Cluster to do data collection of previous migrations statistics, train the machine learning models and then predict live migration cost. Testing results show how the proposed framework can predict live migration time, network throughput and power consumption cost with accurate results and for different kinds of workloads. This helps datacenters admins to have better planning for their VMware environments live migrations.

High-dimensional data is particularly useful for data analytics research. In the healthcare domain, for instance, high-dimensional data analytics has been used successfully for drug discovery. Yet, in order to adhere to privacy legislation, data analytics service providers must guarantee anonymity for data owners. In the context of high-dimensional data, ensuring privacy is challenging because increased data dimensionality must be matched by an exponential growth in the size of the data to avoid sparse datasets. Syntactically, anonymising sparse datasets with methods that rely of statistical significance, makes obtaining sound and reliable results, a challenge. As such, strong privacy is only achievable at the cost of high information loss, rendering the data unusable for data analytics. In this paper, we make two contributions to addressing this problem from both the privacy and information loss perspectives. First, we show that by identifying dependencies between attribute subsets we can eliminate privacy violating attributes from the anonymised dataset. Second, to minimise information loss, we employ a greedy search algorithm to determine and eliminate maximal partial unique attribute combinations. Thus, one only needs to find the minimal set of identifying attributes to prevent re-identification. Experiments on a health cloud based on the SAP HANA platform using a semi-synthetic medical history dataset comprised of 109 attributes, demonstrate the effectiveness of our approach.

This paper explores the development of perceived learner skill confidence in a programming MOOC by applying and analyzing a new Skill Confidence Rating (SCR) survey. After cleaning datasets, we analyze a sample of n = 1689 for the first course module and n = 1147 for the second course module. Results show that on average, learners perceive their skills more confidently after taking a module. The initial confidence per module differs. We could not find a correlation between perceived learner confidence and learner performance in this course.

Mobile devices are omnipresent in our daily lives. They are utilized for a variety of tasks and used multiple times for short periods throughout the day. MOOC providers optimized their platforms for these devices in order to support ubiquitous learning. While a combination of desktop and mobile learning yields improved course performances, standalone learning on mobile devices does not perform in the same manner. One indicator for this is the mismatch between the average usage pattern of mobile devices and the time to consume one content item in a MOOC. Micro learning builds on bite-sized learning material and focusses on short-term learning sessions. This work examines the potential of micro learning activities in the context of MOOCs. Therefore, a framework for video-based micro learning is presented, which features a personalized curriculum. Videos are suggested to the user in a non-linear order that is determined by content dependencies, users' preferences and watched videos, as well as explicit and implicit user feedback. A mobile application was implemented to test the approach with restructured MOOC content resulting in 58 connected short videos about engineering education - e.g. web technologies and programming languages. The usage data indicates initial curiosity by the users. To improve retention rates, more user motivation will be required for future studies. A survey gathered additional qualitative feedback. While the content suggestions were seen as a vital feature for such an approach, the results showed good interest and acceptance rates to create a better learning experience for MOOCs on mobile devices.

The use of digital media and teaching methods pose financial and administrative challenges to many schools. Often educational content is in closed silos, in textbooks or on individual computers in schools. HPI Schul-Cloud (engl. School Cloud) is a German research project funded by the Federal Ministry of Education and Research and realised by the Hasso Plattner Institute for Digital Engineering (HPI) in Potsdam, Germany. The goal of HPI Schul-Cloud is offering a common teaching and learning platform, where teachers are able to plan their lessons and find supportive educational material. Learners can deepen their knowledge using tools for practicing, communication and information retrieval. In this paper we will focus on the material search and discuss the required steps for bringing learning and teaching content into HPI Schul-Cloud.

One of the challenges of educating the next generation of computer scientists is to teach them to become team players, that are able to communicate and interact not only with different IT systems, but also with coworkers and customers with a non-it background. The “bachelor project” is a project based on team work and a close collaboration with selected industry partners. The authors hosted some of the teams since spring term 2014/15. In the paper at hand we explain and discuss this concept and evaluate its success based on students' evaluation and reports. Furthermore, the technology-stack that has been used by the teams is evaluated to understand how self-organized students in IT-related projects work. We will show that and why the bachelor is the most successful educational format in the perception of the students and how this positive results can be improved by the mentors.

Today, Web Analytics (WA) is commonly used to obtain key information about users and their behavior on websites. Besides, with the rise of online learning, Learning Analytics (LA) emerged as a separate research field for collecting and analyzing learners' interactions on online learning platforms. Although the foundation of both methods is similar, WA has not been profoundly used for LA purposes. However, especially large-scale online learning environments may benefit from WA as it is more sophisticated and well-established in comparison to LA. Therefore, this paper aims to examine to what extent WA can be utilized in this context, without compromising the learners' data privacy. For this purpose, Google Analytics was integrated into the Massive Open Online Course platform of the Hasso Plattner Institute as a proof of concept. It was tested with two deployments of the platform: openHPI and openSAP, where thousands of learners gain academic and industry knowledge about engineering education. Besides capturing behavioral data, the platforms' existing LA dashboards were extended by WA metrics. The evaluation of the integration showed that WA covers a large part of the relevant metrics and is particularly suitable for obtaining an overview of the platform's global activity, but reaches its limitations when it comes to learner-specific metrics.

Computer science education in German schools is often less than optimal. It is only mandatory in a few of the federal states and there is a lack of qualified teachers. As a MOOC (Massive Open Online Course) provider with a German background, we developed the idea to implement a MOOC addressing pupils in secondary schools to fill this gap. The course targeted high school pupils and enabled them to learn the Python programming language. In 2014, we successfully conducted the first iteration of this MOOC with more than 7000 participants. However, the share of pupils in the course was not quite satisfactory. So we conducted several workshops with teachers to find out why they had not used the course to the extent that we had imagined. The paper at hand explores and discusses the steps we have taken in the following years as a result of these workshops.

This Research to Practice Full Paper presents the results of a survey among the participants of our MOOC platform about the benefits of what they have learned in our courses and the benefits of the certificates that they have earned for their daily life. We often hear about the high dropout rates in Massive Open Online Courses (MOOCs). On the other hand, there is a lot of movement towards micro-credentials, online master's degrees based on MOOCs, and formal MOOC degrees. However, what is the classic MOOC clientele actually doing with their certificates? What are the reasons why successful MOOC learners put a lot of time and effort in exams and exercises? Do employers accept these certificates in application portfolios? Do they allow their employees to participate in a MOOC during their working hours? Do they pay for the certificates? Are there any differences concerning gender because women are still in the minority in science, technology, engineering and mathematics (STEM) and their career paths often hit a plateau? These questions have been on our mind since we started offering courses on our MOOC platform in 2012. We have had anecdotal evidence, for example, that a participant was helped to get a new job with a certificate from one of our courses-yet this is ultimately just hearsay. Therefore, to find out what is really going on we conducted a survey among the 187,000 registered users of our platform.

Videos are highly instrumental to the learning experience in many eLearning platforms. Lecture videos have become crucial to the success of any Massive Open Online Courses (MOOCs). However, conventional lecture videos are typically large computer files. Such videos are relatively enormous in file-size compared to other learning materials in MOOCs. With a slightly long runtime, a high-definition video file transfer comes with a high price. It is rather much expensive and time-consuming over poor or sparse networks. This commonly occurring situation significantly affects the learning experience. Online learners encounter intermittent connectivity or even offline more often during a learning session. In such situations, we claim that the prevalence of videos in MOOCs remains vague and subtle in poor network environments. In the file transfer process, it is not unusual for a video to face constant network disruptions. Therefore, this research work urges to review the provision of alternative, but more effective multimedia. Such multimedia are instead proposed to be much smaller in file-size. However, the multimedia should retain the essential criterion of a video-based format. The new materials are thus deemed less dependent on the bandwidth. This paper focuses on improvising video-based materials during a real-time learning session. Of course, be derived from their original lecture videos. We argue that this approach helps to improve the learning experience, especially for learners situated in low bandwidth areas. Network related issues such as long-wait buffering, low-quality streaming, or slow downloading are therefore tackled by using unconventional materials. The main benefit is to achieve and attain uninterrupted learning session even when the MOOC learner steps into offline or out of reach area. Also, the approach supplements images and audio to support sustainable eLearning – whenever wherever.

Binary neural networks are a promising approach to execute convolutional neural networks on devices with low computational power. Previous work on this subject often quantizes pretrained full-precision models and uses complex training strategies. In our work, we focus on increasing the performance of binary neural networks by training from scratch with a simple training strategy. In our experiments we show that we are able to achieve state-of-the-art results on standard benchmark datasets. Further, we analyze how full-precision network structures can be adapted for efficient binary networks and adopt a network architecture based on a DenseNet for binary networks, which lets us improve the state-of-the-art even further. Our source code can be found online: https://github.com/hpi-xnor/BMXNet-v2.

Modern information infrastructures and organizations increasingly face the problem of data breaches and cyber-attacks. A traditional method for dealing with this problem are classification zones, such as ‘top secret’, ‘confidential’, and ‘unclassified’, which regulate the access of persons, hardware, and software to data records. In this paper, we present an approach that finds classification zone violations through automated message flow analysis. Our approach considers the problem of anonymization for the source event logs, which makes the resulting data flow model sharable with experts and the public. We discuss practical implications from applying the approach to a large governmental organization data set and discuss how the anonymity of our concept can be formally validated.

With Massive Open Online Courses (MOOCs) the number of people having access to higher education increased rapidly. The intentions to enroll for a specific course vary significantly and depend on one's professional or personal learning needs and interests. All learners have in common that they pursue their individual learning objectives. However, predominant MOOC platforms follow a one-size-fits-all approach and primarily aim for completion with certification. Specifically, technical support for goal-oriented and self-regulated learning to date is very limited in this context although both learning strategies are proven to be key factors for students' achievement in large-scale online learning environments. In this first investigation, a concept for the application and technical integration of personalized learning objectives in a MOOC platform is realized and assessed. It is evaluated with a mixed-method approach. First, the learners' acceptance is examined with a multivariate A/B test in two courses. Second, a survey was conducted to gather further feed-back about the perceived usefulness, next to the acceptance. The results show a positive perception by the learners, which paves the way for future research.

The ability to work in teams is an important skill in today's work environments. In MOOCs, however, team work, team tasks, and graded team-based assignments play only a marginal role. To close this gap, we have been exploring ways to integrate graded team-based assignments in MOOCs. Some goals of our work are to determine simple criteria to match teams in a volatile environment and to enable a frictionless online collaboration for the participants within our MOOC platform. The high dropout rates in MOOCs pose particular challenges for team work in this context. By now, we have conducted 15 MOOCs containing graded team-based assignments in a variety of topics. The paper at hand presents a study that aims to establish a solid understanding of the participants in the team tasks. Furthermore, we attempt to determine which team compositions are particularly successful. Finally, we examine how several modifications to our platform's collaborative toolset have affected the dropout rates and performance of the teams.

The self-sovereign identity management model emerged with the rise of blockchain technology. This paradigm focuses on user-centricity and strives to place the user in full control of the digital identity. Numerous implementations embrace the self-sovereign identity concept, leading to a fragmented landscape of solutions. At the same time, traditional identity and access management protocols are largely disregarded and facilities to issue verifiable claims as attributes are not available. Therefore, service providers barely adopt these solutions. We propose a component-based architecture for integrating self-sovereign identity solutions into web applications to foster their adoption by service providers. Furthermore, we outline a sample implementation as a gateway that enables uPort and Jolocom for authentication, via the OpenID Connect protocol, as well as the retrieval of email address attestations for these solutions.

Cyber-attacks against cloud storage infrastructure e.g. Amazon S3 and Google Cloud Storage, have increased in recent years. One reason for this development is the rising adoption of cloud storage for various purposes. Robust counter-measures are therefore required to tackle these attacks especially as traditional techniques are not appropriate for the evolving attacks. We propose a two-pronged approach to address these challenges in this paper. The first approach involves dynamic snapshotting and recovery strategies to detect and partially neutralize security events. The second approach builds on the initial step by automatically correlating the generated alerts with cloud event log, to extract actionable intelligence for incident response. Thus, malicious activities are investigated, identified and eliminated. This approach is implemented in SlingShot, a cloud threat detection and incident response system which extends our earlier work - CSBAuditor, which implements the first step. The proposed techniques work together in near real time to mitigate the aforementioned security issues on Amazon Web Services (AWS) and Google Cloud Platform (GCP). We evaluated our techniques using real cloud attacks implemented with static and dynamic methods. The average Mean Time to Detect is 30 seconds for both providers, while the Mean Time to Respond is 25 minutes and 90 minutes for AWS and GCP respectively. Thus, our proposal effectively tackles contemporary cloud attacks.

The majority of security breaches in cloud infrastructure in recent years are caused by human errors and misconfigured resources. Novel security models are imperative to overcome these issues. Such models must be customer-centric, continuous, not focused on traditional security paradigms like intrusion detection and adopt proactive techniques. Thus, this paper proposes CloudStrike, a cloud security system that implements the principles of Chaos Engineering to enable the aforementioned properties. Chaos Engineering is an emerging discipline employed to prevent non-security failures in cloud infrastructure via Fault Injection Testing techniques. CloudStrike employs similar techniques with a focus on injecting failures that impact security i.e. integrity, confidentiality and availability. Essentially, CloudStrike leverages the relationship between dependability and security models. Preliminary experiments provide insightful and prospective results.

We propose a new generative adversarial architecture to mitigate imbalance data problem in medical image semantic segmentation where the majority of pixels belongs to a healthy region and few belong to lesion or non-health region. A model trained with imbalanced data tends to bias towards healthy data which is not desired in clinical applications and predicted outputs by these networks have high precision and low sensitivity. We propose a new conditional generative refinement network with three components: a generative, a discriminative, and a refinement networks to mitigate imbalanced data problem through ensemble learning. The generative network learns to the segment at the pixel level by getting feedback from the discriminative network according to the true positive and true negative maps. On the other hand, the refinement network learns to predict the false positive and the false negative masks produced by the generative network that has significant value, especially in medical application. The final semantic segmentation masks are then composed by the output of the three networks. The proposed architecture shows state-of-the-art results on LiTS-2017 for simultaneous liver and lesion segmentation, and MDA231 for microscopic cell segmentation. We have achieved competitive results on BraTS-2017 for brain tumor segmentation.

Detecting and recognizing text in natural scene images is a challenging, yet not completely solved task. In recent years several new systems that try to solve at least one of the two sub-tasks (text detection and text recognition) have been proposed. In this paper we present SEE, a step towards semi-supervised neural networks for scene text detection and recognition, that can be optimized end-to-end. Most existing works consist of multiple deep neural networks and several pre-processing steps. In contrast to this, we propose to use a single deep neural network, that learns to detect and recognize text from natural images, in a semi-supervised way. SEE is a network that integrates and jointly learns a spatial transformer network, which can learn to detect text regions in an image, and a text recognition network that takes the identified text regions and recognizes their textual content. We introduce the idea behind our novel approach and show its feasibility, by performing a range of experiments on standard benchmark datasets, where we achieve competitive results.

Cloud storage brokerage is an abstraction aimed at providing value-added services. However, Cloud Service Brokers are challenged by several security issues including enlarged attack surfaces due to integration of disparate components and API interoperability issues. Therefore, appropriate security risk assessment methods are required to identify and evaluate these security issues, and examine the efficiency of countermeasures. A possible approach for satisfying these requirements is employment of threat modeling concepts, which have been successfully applied in traditional paradigms. In this work, we employ threat models including attack trees, attack graphs and Data Flow Diagrams against a Cloud Service Broker (CloudRAID) and analyze these security threats and risks. Furthermore, we propose an innovative technique for combining Common Vulnerability Scoring System (CVSS) and Common Configuration Scoring System (CCSS) base scores in probabilistic attack graphs to cater for configuration-based vulnerabilities which are typically leveraged for attacking cloud storage systems. This approach is necessary since existing schemes do not provide sufficient security metrics, which are imperatives for comprehensive risk assessments. We demonstrate the efficiency of our proposal by devising CCSS base scores for two common attacks against cloud storage: Cloud Storage Enumeration Attack and Cloud Storage Exploitation Attack. These metrics are then used in Attack Graph Metric-based risk assessment. Our experimental evaluation shows that our approach caters for the aforementioned gaps and provides efficient security hardening options. Therefore, our proposals can be employed to improve cloud security.

The detection of all inclusion dependencies (INDs) in an unknown dataset is at the core of any data profiling effort. Apart from the discovery of foreign key relationships, INDs can help perform data integration, integrity checking, schema (re-)design, and query optimization. With the advent of Big Data, the demand increases for efficient INDs discovery algorithms that can scale with the input data size. To this end, we propose S-indd++ as a scalable system for detecting unary INDs in large datasets. S-indd++ applies a new stepwise partitioning technique that helps discard a large number of attributes in early phases of the detection by processing the first partitions of smaller sizes. S-indd++ also extends the concept of the attribute clustering to decide which attributes to be discarded based on the clustering result of each partition. Moreover, in contrast to the state-of-the-art, S-indd++ does not require the partition to fit into the main memory- which is a highly appreciable property in the face of the ever growing datasets. We conducted an exhaustive evaluation of S-indd ++ by applying it to large datasets with thousands attributes and more than 266 million tuples. The results show the high superiority of S-indd++ over the state-of-the-art. S-indd++ reduced up to 50~% of the runtime in comparison with Binder, and up to 98~% in comparison with S-indd.

In university teaching today, it is common practice to record regular lectures and special events such as conferences and speeches. With these recordings, a large fundus of video teaching material can be created quickly and easily. Typically, lectures have a length of about one and a half hours and usually take place once or twice a week based on the credit hours. Depending on the number of lectures and other events recorded, the number of recordings available is increasing rapidly, which means that an appropriate form of provisioning is essential for the students. This is usually done in the form of lecture video platforms. In this work, we have investigated how lecture video platforms and the contained knowledge can be improved and accessed more easily by an increasing number of students. We came up with a multistep process we have applied to our own lecture video web portal that can be applied to other solutions as well.

When students watch learning videos online, they usually need to watch several hours of video content. In the end, not every minute of a video is relevant for the exam. Additionally, students need to add notes to clarify issues of a lecture. There are several possibilities to enhance the metadata of a video, e.g. a typical way to add user-specific information to an online video is a comment functionality, which allows users to share their thoughts and questions with the public. In contrast to common video material which can be found online, lecture videos are used for exam preparation. Due to this difference, the idea comes up to annotate lecture videos with markers and personal notes for a better understanding of the taught content. Especially, students learning for an exam use their notes to refresh their memories. To ease this learning method with lecture videos, we introduce the annotation feature in our video lecture archive. This functionality supports the students with keeping track of their thoughts by providing an intuitive interface to easily add, modify or remove their ideas. This annotation function is integrated in the video player. Hence, scrolling to a separate annotation area on the website is not necessary. Furthermore, the annotated notes can be exported together with the slide content to a PDF file, which can then be printed easily. Lecture video annotations support and motivate students to learn and watch videos from an E-Learning video archive.

MOOCs (Massive Open Online Courses) become more and more popular for learners of all ages to study further or to learn new subjects of interest. The purpose of this paper is to introduce a different MOOC course style. Typically, video content is shown teaching the student new information. After watching a video, self-test questions can be answered. Finally, the student answers weekly exams and final exams like the self-test questions. Out of the points that have been scored for weekly and final exams a certificate can be issued. Our approach extends the possibility to receive points for the final score with practical programming exercises on real hardware. It allows the student to do embedded programming by communicating over GPIO pins to control LEDs and measure sensor values. Additionally, they can visualize values on an embedded display using web technologies, which are an essential part of embedded and smart home devices to communicate with common APIs. Students have the opportunity to solve all tasks within the online remote lab and at home on the same kind of hardware. The evaluation of this MOOCs indicates the interesting design for students to learn an engineering technique with new technology approaches in an appropriate, modern, supporting and motivating way of teaching.

This Research-to-Practice paper examines the practical application of various forms of collaborative learning in MOOCs. Since 2012, about 60 MOOCs in the wider context of Information Technology and Computer Science have been conducted on our self-developed MOOC platform. The platform is also used by several customers, who either run their own platform instances or use our white label platform. We, as well as some of our partners, have experimented with different approaches in collaborative learning in these courses. Based on the results of early experiments, surveys amongst our participants, and requests by our business partners we have integrated several options to offer forms of collaborative learning to the system. The results of our experiments are directly fed back to the platform development, allowing to fine tune existing and to add new tools where necessary. In the paper at hand, we discuss the benefits and disadvantages of decisions in the design of a MOOC with regard to the various forms of collaborative learning. While the focus of the paper at hand is on forms of large group collaboration, two types of small group collaboration on our platforms are briefly introduced.

In the course of patient treatments, psychotherapists aim to meet the challenges of being both a trusted, knowledgeable conversation partner and a diligent documentalist. We are developing the digital whiteboard system Tele-Board MED (TBM), which allows the therapist to take digital notes during the session together with the patient. This study investigates what therapists are experiencing when they document with TBM in patient sessions for the first time and whether this documentation saves them time when writing official clinical documents. As the core of this study, we conducted four anamnesis session dialogues with behavior psychotherapists and volunteers acting in the role of patients. Following a mixed-method approach, the data collection and analysis involved self-reported emotion samples, user experience curves and questionnaires. We found that even in the very first patient session with TBM, therapists come to feel comfortable, develop a positive feeling and can concentrate on the patient. Regarding administrative documentation tasks, we found with the TBM report generation feature the therapists save 60% of the time they normally spend on writing case reports to the health insurance.

Automatic tumor segmentation is an important and challenging clinical task because tumors have different sizes, shapes, contrasts, and locations. In this paper, we present an automatic instance semantic segmentation method based on deep neural networks (DNNs). The proposed networks are tailored to picture tumors in magnetic resonance imaging (MRI) and computed tomography (CT) images. We present an end-to-end multitask learning architecture comprising three stages, namely detection, segmentation, and classification. This paper introduces a new technique for tumor detection based on high-level extracted features from convolutional neural networks (CNNs) using the Hough transform technique. The detected tumor(s), are segmented with a set of fully connected (FC) layers, and the segmented mask is classified through FCs. The proposed architecture gives promising results on the popular medical image benchmarks. Our framework is generalized in the sense that it can be used in different types of medical images in varied sizes, such as the Liver Tumor Segmentation (LiTS-2017) challenge, and the Brain Tumor Segmentation (BraTS-2016) benchmark.

Teamwork and collaborative learning are considered superior to learning individually by many instructors and didactical theories. Particularly, in the context of e-learning and Massive Open Online Courses (MOOCs) we see great benefits but also great challenges for both, learners and instructors. We discuss our experience with six team based assignments on the openHPI and openSAP1 MOOC platforms.

Cloud Storage Brokers (CSB) provide seamless and concurrent access to multiple Cloud Storage Services (CSS) while abstracting cloud complexities from end-users. However, this multi-cloud strategy faces several security challenges including enlarged attack surfaces, malicious insider threats, security complexities due to integration of disparate components and API interoperability issues. Novel security approaches are imperative to tackle these security issues. Therefore, this paper proposes CS-BAuditor, a novel cloud security system that continuously audits CSB resources, to detect malicious activities and unauthorized changes e.g. bucket policy misconfigurations, and remediates these anomalies. The cloud state is maintained via a continuous snapshotting mechanism thereby ensuring fault tolerance. We adopt the principles of chaos engineering by integrating BrokerMonkey, a component that continuously injects failure into our reference CSB system, CloudRAID. Hence, CSBAuditor is continuously tested for efficiency i.e. its ability to detect the changes injected by BrokerMonkey. CSBAuditor employs security metrics for risk analysis by computing severity scores for detected vulnerabilities using the Common Configuration Scoring System, thereby overcoming the limitation of insufficient security metrics in existing cloud auditing schemes. CSBAuditor has been tested using various strategies including chaos engineering failure injection strategies. Our experimental evaluation validates the efficiency of our approach against the aforementioned security issues with a detection and recovery rate of over 96 %.

Cloud storage brokerage systems abstract cloud storage complexities by mediating technical and business relationships between cloud stakeholders, while providing value-added services. This however raises security challenges pertaining to the integration of disparate components with sometimes conflicting security policies and architectural complexities. Assessing the security risks of these challenges is therefore important for Cloud Storage Brokers (CSBs). In this paper, we present a threat modeling schema to analyze and identify threats and risks in cloud brokerage brokerage systems. Our threat modeling schema works by generating attack trees, attack graphs, and data flow diagrams that represent the interconnections between identified security risks. Our proof-of-concept implementation employs the Common Configuration Scoring System (CCSS) to support the threat modeling schema, since current schemes lack sufficient security metrics which are imperatives for comprehensive risk assessments. We demonstrate the efficiency of our proposal by devising CCSS base scores for two attacks commonly launched against cloud storage systems: Cloud sStorage Enumeration Attack and Cloud Storage Exploitation Attack. These metrics are then combined with CVSS based metrics to assign probabilities in an Attack Tree. Thus, we show the possibility combining CVSS and CCSS for comprehensive threat modeling, and also show that our schemas can be used to improve cloud security.

User-generated content on social media platforms is a rich source of latent information about individual variables. Crawling and analyzing this content provides a new approach for enterprises to personalize services and put forward product recommendations. In the past few years, brands made a gradual appearance on social media platforms for advertisement, customers support and public relation purposes and by now it became a necessity throughout all branches. This online identity can be represented as a brand personality that reflects how a brand is perceived by its customers. We exploited recent research in text analysis and personality detection to build an automatic brand personality prediction model on top of the (Five-Factor Model) and (Linguistic Inquiry and Word Count) features extracted from publicly available benchmarks. The proposed model reported significant accuracy in predicting specific personality traits form brands. For evaluating our prediction results on actual brands, we crawled the Facebook API for 100k posts from the most valuable brands' pages in the USA and we visualize exemplars of comparison results and present suggestions for future directions.

This paper introduces VuWaDB, which is a database for vulnerability workarounds. When faced with a newly discovered vulnerability, experts tend to use workarounds in order to mitigate the risks until a patch is issued by the vendor. Currently, the available vulnerability databases suffer from the lack of comprehensive workaround solutions. Furthermore, the presented workarounds are only limited to a few vulnerabilities but also poorly recorded in natural language sentences and plain text format. In order to construct VuWaDB, first, the related information was gathered from a number of well-known vulnerability databases and then extracted, analyzed, and labeled. In this regard, VuWaDB organizes the workarounds in six categories: configuration, modify code, redirect, remove, restrict access and, utility tool. Lastly, it was analyzed from the statistical point of view.

Generating a novel and descriptive caption of an image is drawing increasing interests in computer vision, natural language processing, and multimedia communities. In this work, we propose an end-to-end trainable deep bidirectional LSTM (Bi-LSTM (Long Short-Term Memory)) model to address the problem. By combining a deep convolutional neural network (CNN) and two separate LSTM networks, our model is capable of learning long-term visual-language interactions by making use of history and future context information at high-level semantic space. We also explore deep multimodal bidirectional models, in which we increase the depth of nonlinearity transition in different ways to learn hierarchical visual-language embeddings. Data augmentation techniques such as multi-crop, multi-scale, and vertical mirror are proposed to prevent overfitting in training deep models. To understand how our models “translate” image to sentence, we visualize and qualitatively analyze the evolution of Bi-LSTM internal states over time. The effectiveness and generality of proposed models are evaluated on four benchmark datasets: Flickr8K, Flickr30K, MSCOCO, and Pascal1K datasets. We demonstrate that Bi-LSTM models achieve highly competitive performance on both caption generation and image-sentence retrieval even without integrating an additional mechanism (e.g., object detection, attention model). Our experiments also prove that multi-task learning is beneficial to increase model generality and gain performance. We also demonstrate the performance of transfer learning of the Bi-LSTM model significantly outperforms previous methods on the Pascal1K dataset.

Micro-grids offer a cost-effective approach to providing reliable power supply in isolated and disadvantaged communities. These communities present a special case where access to national power networks is either non-existent or intermittent due to load-shedding to provision urban areas and/or due to high interconnection costs. By necessity, such micro-grids rely on renewable energy sources that are variable and so only partly predictable. Ensuring reliable power provisioning and billing must therefore be supported by demand management and fair-billing policies. Furthermore, since trusted centralized grid management is not always possible, using a distributed model offers a viable solution approach. However, such a distributed system may be subject to subversion attacks aimed at power theft. In this paper, we present a novel and innovative distributed architecture for power distribution and billing on micro-grids. The architecture is designed to operate efficiently over a lossy communication network, which is an advantage for disadvantaged communities. Since lossy networks are undependable, differentiating system failures from adversarial manipulations is important because grid stability is to a large extent dependent on user participation. To this end, we provide a characterization of potential adversarial models to underline how these can be differentiated from failures.

The number of attacks that use sophisticated and complex methods increased lately. The main objective of these attacks is to largely infiltrate the target network and to stay undetected. Therefore, the attackers often use valid credentials and standard administrative tools to hide between legitimate user actions and to hinder detection. Most existing security systems, which use standard signature-based or anomaly-based approaches, are not able to identify this type of malicious activities. Furthermore, it is also most often not feasible to analyze user behavior manually, due to the complexity of this task and the high amount of different user actions. Thus, it is necessary to develop new automated approaches to identify suspicious user behavior. In this paper, we propose to use neural networks to analyze user behavior and to identify suspicious actions. Due to the fact that neural networks require suitable datasets to learn the difference between suspicious and benign actions, we describe a behavioral simulation system to generate reasonable datasets. These datasets use different behavioral features to describe log-on and log-off activities of users. To identify suitable neural network models for user behavior analysis, we evaluate and compare 16,275 different feed-forward neural networks with three different datasets and 75 recurrent neural networks with one dataset. The results show that the used dataset and the complexity of a model are crucial to achieve a high accuracy. Appropriate models, which also consider context behavior information, are able to automatically classify before unseen user actions with an accuracy of up to 98 %.

Nowadays, social networks are an essential part of modern life and they are being adopted as an additional channel for informing people about extreme events and incidents. Emergency agencies actively use the potential of the social networks from the perspective of providing situational and public safety awareness. One of such agencies is the Berlin Police that is the subject of our interest in this paper. They use its Twitter account to inform Berlin inhabitants about incidents that happen in the city. Also, they provide the Twitter marathons and campaigns to demonstrate how the police work and to motivate users to support the police by posting tweets about incidents if they became witnesses. In this paper, we analyze data from the Twitter account of the Berlin Police to understand the public safety situation in the city. We provide the analysis of the entire gathered data and data of two police Twitter marathons. Our analysis results show the different views on the public safety situation in the city: distribution of incidents in districts of the city, distribution of different types of incidents, the intensity of the incidents during the day and others.

A wide range of Massive Open Online Courses (MOOCs) on various topics has been offered since 2008. The high expectations of this new way of teaching, however, were not entirely met until now due to different reasons. To better engage participants in MOOCs, gamification was implemented on many platforms; especially since gamification became a buzz-word in 2012. Gamification, however, did not stop people from dropping out of courses as efficiently as expected. Although completion rates are no longer assumed to be the only evidence for the users' success, different approaches to improve both the learners' experience and motivation should be evaluated to increase the advantages of MOOCs. In the scope of gamified learning, the use of Digital Game-Based Learning (DGBL) to offer learning in an even more playful way than with gamification is worth a closer look. In this paper, the current implementation of gamification, DGBL and their role within MOOCs will be evaluated. It is also explained why and how playing can be beneficial for learning. Possible approaches how to integrate DGBL in MOOCs as well as advantages, disadvantages, and different needs will be discussed. A survey will be presented that gives a first hint about the participants' desires and needs. In conclusion, the paper presents current as well as possible future research on how to best integrate game-based learning into MOOCs for engineering education, specifically at the openHPI platform.

Massive Open Online Courses (MOOCs) have left their mark on the face of education during the recent years. At the Hasso Plattner Institute (HPI) in Potsdam, Germany, we are actively developing a MOOC platform, which provides our research with a plethora of e-learning topics, such as learning analytics, automated assessment, peer assessment, team-work, online proctoring, and gamification. We run several instances of this platform. On openHPI, we provide our own courses from within the HPI context. Further instances are openSAP, openWHO, and mooc.HOUSE, which is the smallest of these platforms, targeting customers with a less extensive course portfolio. In 2013, we started to work on the gamification of our platform. By now, we have implemented about two thirds of the features that we initially have evaluated as useful for our purposes. About a year ago we activated the implemented gamification features on mooc.HOUSE. Before activating the features on openHPI as well, we examined, and re-evaluated our initial considerations based on the data we collected so far and the changes in other contexts of our platforms.

In this paper we propose an integrated framework of automatic bilingual subtitle generation for lecture videos, especially for MOOCs. The framework consists of Automatic Speech Recognition (ASR), Sentence Boundary Detection (SBD), and Machine Translation (MT). Then we quantitatively evaluate the auto-generated subtitles, the manually produced subtitles from scratch, and the auto-generated subtitles with manual modification in term of accuracy and time expenditure, in both original and target languages. The result shows that the auto-generated subtitles in the original language (English) are fairly accurate already. By using them as the draft, human subtitle producers can save 54% of the working time and simultaneously reduce the error rate by 54.3%, which is a significant improvement. However, the effectiveness of machine translated subtitles (English to Chinese) is limited. In the end, if the proposed framework is applied, the total working time in preparing bilingual subtitles can be shortened by approximately 1/3, with no decline in quality.

Tools to generate game-based learning materials are seldom implemented in Massive Open Online Course (MOOC) platforms. First experiments with game-based learning MOOCs had shown good results. In my research, I will focus on the integration of game-based learning to the openHPI MOOC platform. The use of game-based learning materials will be evaluated in three iterations: (1) by existing means the platform already offers, (2) by simple interactive exercises, and (3) by an educational game. The success of all iterations will be evaluated with user surveys and A/B testing. The research goal is to establish a tool for creating interactive exercises as well as the design of an appropriate educational game. Advantages and disadvantages of different game-based methods as well as best practices will be listed in the process.

Generating seeds on Internet of things (IoT) devices is challenging because these devices typically lack common entropy sources, such as user interaction or hard disks. A promising replacement is to use power-up static random-access memory (SRAM) states, which are partly random due to manufacturing deviations. Thus far, there, however, seems to be no method for extracting close-to-uniformly distributed seeds from power-up SRAM states in an information-theoretically secure and practical manner. Moreover, the min-entropy of power-up SRAM states reduces with temperature, thereby rendering this entropy source vulnerable to so-called freezing attacks. In this paper, we mainly make three contributions. First, we propose a new method for extracting uniformly distributed seeds from power-up SRAM states. Unlike current methods, ours is information-theoretically secure, practical, and freezing attack-resistant rolled into one. Second, we point out a trick that enables using power-up SRAM states not only for self-seeding at boot time, but also for reseeding at runtime. Third, we compare the energy consumption of seeding an IoT device either with radio noise or power-up SRAM states. While seeding with power-up SRAM states turned out to be more energy efficient, we argue for mixing both these entropy sources.

Teamwork is an an important topic in education. It fosters deep learning and allows educators to assign interesting tasks, which would be too complex to be solved by single participants due to the time restrictions defined by the context of a course.Furthermore, today's jobs require an increasing amount of team skills. On the other hand, teamwork comes with a variety of issues of its own. Particularly in large scale settings, such as MOOCs, teamwork is challenging. Courses often end with dysfunctional teams due to drop-outs or insufficient matching. The paper at hand presents a set of three tools that we have recently added to our system to enable teamwork in our courses. This toolset consists of the TeamBuilder, a tool to match successful teams based on a variable set of parameters, CollabSpaces, providing teams with a secluded area to communicate and collaborate within the course context, and a TeamPeerAssessment tool, which allows to provide teams with complex tasks and which allows assessment that suffiiently scales for the MOOC context. The presented tools are evaluated in terms of success rates of the created teams and workload reduction for the courses' teaching teams.

Binary Neural Networks (BNNs) can drastically reduce memory size and accesses by applying bit-wise operations instead of standard arithmetic operations. Therefore it could significantly improve the efficiency and lower the energy consumption at runtime, which enables the application of state-of-the-art deep learning models on low power devices. BMXNet is an open-source BNN library based on MXNet, which supports both XNOR-Networks and Quantized Neural Networks. The developed BNN layers can be seamlessly applied with other standard library components and work in both GPU and CPU mode. BMXNet is maintained and developed by the multimedia research group at Hasso Plattner Institute and released under Apache license. Extensive experiments validate the efficiency and effectiveness of our implementation. The BMXNet library, several sample projects, and a collection of pre-trained binary deep models are available for download at https://github.com/hpi-xnor.

IoT devices usually are battery-powered and directly connected to the Internet. This makes them vulnerable to so-called path-based denial-of-service (PDoS) attacks. For example, in a PDoS attack an adversary sends multiple Constrained Application Protocol (CoAP) messages towards an IoT device, thereby causing each IoT device along the path to expend energy for forwarding this message. Current end-to-end security solutions, such as DTLS or IPsec, fail to prevent such attacks since they only filter out inauthentic CoAP messages at their destination. This demonstration shows an approach to allow en-route filtering where a trusted gateway has all necessary information to check the integrity, decrypt and, if necessary, drop a message before forwarding it to the constrained mote. Our approach preserves precious resources of IoT devices in the face of path-based denial-of-service attacks by remote attackers.

CloudRAID is a secure personal cloud storage broker that provides data availability, security, and privacy for private usage. But some of the challenges need to be resolved to use CloudRAID as an enterprise cloud storage broker solution, such as complicated key management, absence of role-based hierarchical access control, and lack of administrative oversight. In this paper we tackle these challenges and propose an enterprise version of CloudRAID called CloudRAID for Business (CfB). We combine CloudRAID with Ciphertext-Policy Attribute-Based Encryption (CP-ABE) and implement administrative oversight for monitoring activities in CfB system and multiple CSPs. Our evaluation of CfB demonstrates that it offers robust security measures through fine-grained role-based access control, scalable key management for multi-user-and-device scenarios, reduces complexity of file sharing revocation, file-level security, and administrative oversight.

Digital organizations are now ruled by the flexibility offered by remote working environments. The availability of people to work from different places, with digital tools, empower them very often not only to perform their work better and faster, but also to reach higher levels of satisfaction in balancing work and private life. However, from an organizational perspective, there are concerns regarding the management of knowledge of those remote individuals and how they may impact the further creation of knowledge within the organization. Considering that knowledge is not only information stored on databases or written in documents, the interaction between peers is also an enormous source of knowledge that needs to be canalized. Commonly described as tacit knowledge, face-to-face communication is the main enabler of such knowledge and is one is the missing elements in remote working environments. For a geographically distributed team, a common `virtual space', personalized to the cognitive needs of each individual, would represent an improvement in the digital space and could compensate for the missing face-to-face interaction. This paper proposes integrating cognitive adaptability in virtual working environments to enhance user interaction and knowledge processing, taking into consideration behavioral conditions and individuals' learning diversity. Since interaction in decentralized environments is ruled by online behaviors, the use of virtual reality and the sensorimotor experience it offers is proposed to design individual virtual spaces for interaction, which should be created as closely as possible to their ideal learning dimension. Designing individualized environments focusses on creating optimal surrounding digital spaces that precisely match the cognitive and learning conditions of individuals, so that interaction and knowledge creation are not compromised, but improved.

Inclusion dependencies form one of the most fundamental classes of integrity constraints. Their importance in classical data management is reinforced by modern applications such as data profiling, data cleaning, entity resolution and schema matching. Their discovery in an unknown dataset is at the core of any data analysis effort. Therefore, several research approaches have focused on their efficient discovery in a given, static dataset. However, none of these approaches are appropriate for applications on dynamic datasets, such as transactional datasets, scientific applications, and social network. In these cases, discovery techniques should be able to efficiently update the inclusion dependencies after an update in the dataset, without reprocessing the entire dataset.

Summary Modern security information and event management systems should be capable to store and process high amount of events or log messages in different formats and from different sources. This requirement often prevents such systems from usage of computational heavy algorithms for security analysis. To deal with this issue, we built our system based on an in‐memory database with an integrated machine learning library, namely, SAP HANA. Three approaches, that is, (1) deep normalisation of log messages, (2) storing data in the main memory and (3) running data analysis directly in the database, allow us to increase processing speed in such a way that machine learning analysis of security events becomes possible nearly in real time. Besides that, we developed a universal anomaly detection algorithm, which uses vector space model to represent and cluster textual log messages. Together with deep normalisation approach, this algorithm solves the problem of correlation for heterogenous security events containing many text fields. To prove our concepts, we measured the processing speed for the developed system on the data generated using Active Directory testbed, compared it with classical system architecture based on PostgreSQL database and showed the efficiency of our approach for high‐speed analysis of security events. Copyright © 2016 John Wiley & Sons, Ltd.

The practice of rejecting injected and replayed 802.15.4 frames only after they were received leaves 802.15.4 nodes vulnerable to broadcast and droplet attacks. Basically, in broadcast and droplet attacks, an attacker injects or replays plenty of 802.15.4 frames. As a result, victim 802.15.4 nodes stay in receive mode for extended periods of time and expend their limited energy. He et al. considered embedding one-time passwords in the synchronization headers of 802.15.4 frames so as to avoid that 802.15.4 nodes detect injected and replayed 802.15.4 frames in the first place. However, He et al.'s, as well as similar proposals lack support for broadcast frames and depend on special hardware. In this paper, we propose Practical On-the-fly Rejection (POTR) to reject injected and replayed 802.15.4 frames early during receipt. Unlike previous proposals, POTR supports broadcast frames and can be implemented with many off-the-shelf 802.15.4 transceivers. In fact, we implemented POTR with CC2538 transceivers, as well as integrated POTR into the Contiki operating system. Furthermore, we demonstrate that, compared to using no defense, POTR reduces the time that 802.15.4 nodes stay in receive mode upon receiving an injected or replayed 802.15.4 frame by a factor of up to 16. Beyond that, POTR has a small processing and memory overhead, and incurs no communication overhead.

Mitigation techniques employed by attackers has meant that traditional Network Intrusion Detection Systems (NIDS) are no longer able to reliably protect a network in the face of ever more sophisticated attacks. Security Information and Event Management (SIEM) systems monitor network systems by analyzing the logs they produce. In this paper, we propose a method of visualizing attacks by aggregating, normalizing and analyzing alerts raised by SIEM-based IDS (SIDS) systems as well as NIDS systems in real-time. We present the results of our proposed visualization technique when applied to different attack scenarios. In many cases, our approach allows for the path an attacker takes during their attack to be visualized.

Remote collaboration systems are a necessity for geographically dispersed teams in achieving a common goal. Real-time groupware systems frequently provide a shared workspace where users interact with shared artifacts. However, a shared workspace is often not enough for maintaining the awareness of other users. Video conferencing can create a visual context simplifying the user's communication and understanding. In addition, flexible working modes and modern communication systems allow users to work at any time at any location. It is therefore desirable that a groupware system can run on users' everyday devices, such as smartphones and tablets, in the same way as on traditional desktop hardware. We present a standards compliant, web browser-based real-time remote collaboration system that includes WebRTC-based video conferencing. It allows a full-body video setup where everyone can see what other participants are doing and where they are pointing in the shared workspace. In contrast to standard WebRTC's peer-to-peer architecture, our system implements a star topology WebRTC video conferencing. In this way, our solution improves network bandwidth efficiency from a linear to a constant network upstream consumption.

Earlier research shows that using an embedded LED system motivates students to learn programming languages in massive open online courses (MOOCs) efficiently. Since this earlier approach was very successful the system should be improved to increase the learning experience for students during programming exercises. The problem of the current system is that only a static image was shown on the LED matrix controlled by students' array programming over the embedded system. The idea of this paper is to change this static behavior into a dynamic display of information on the LED matrix by the use of sensors which are connected with the embedded system. For this approach a light sensor and a temperature sensor are connected to an analog-to-digital converter (ADC) port of the embedded system. These sensors' values can be read by the students to compute the correct output for the LED matrix. The result is captured and sent back to the students for direct feedback. Furthermore, unit tests can be used to automatically evaluate the programming results. The system was evaluated during a MOOC course about Web Technologies using JavaScript. Evaluation results are taken from the student's feedback and an evaluation of the students' code executions on the system. The positive feedback and the evaluation of the students' executions, which shows a higher amount of code executions compared to standard programming tasks and the fact that students solving these tasks have overall better course results, highlight the advantage of the approach. Due to the evaluation results, this approach should be used in e-learning e.g. MOOCs teaching programming languages to increase the learning experience and motivate students to learn programming.

The paper at hand introduces CodeOcean, a web-based platform to provide practical programming exercises. CodeOcean is designed to be used in Massive Open Online Courses (MOOCs) to teach programming to beginners. Its concept and implementation are discussed with regard to tools provided to students and teachers, sandboxed and scalable code execution, scalable assessment, and interoperability. MOOCs bear a tremendous potential for teaching programming to a large and diverse audience. Learning to program, however, is a hands-on effort; watching videos and solving multiple choice tests will not be sufficient. A platform, such as CodeOcean, to work on practical programming exercises and to solve actual programming tasks is required. Due to the massiveness of the courses, teaching teams cannot check, give feedback, or assess the submissions of the participants manually. CodeOcean provides the participants with proper automated feedback in a timely manner and is able to assess the given programming tasks in an automated way.

Live migration is one of the powerful features in virtual datacenters environment. Servers load balance, power saving and dynamic resource management techniques are all dependent on live migration feature in virtual datacenters. So it is important to study virtual machine live migration processes and analyze its performance impact on datacentres resources. In this research, the performance analysis for single and multiple virtual machines migration has led to getting empirical models that can be used for live migration overhead estimation and providing resource management techniques that are migration overhead aware.

In this paper we propose a method to evaluate the importance of lecture video segments in online courses. The video will be first segmented based on the slide transition. Then we evaluate the importance of each segment based on our analysis of the teacher's focus. This focus is mainly identified by exploring features in the slide and the speech. Since the whole analysis process is based on multimedia materials, it could be done before the official start of the course. By setting survey questions and collecting forum statistics in the MOOC "Web Technologies", the proposed method is evaluated. Both the general trend and the high accuracy of selected key segments (over 70%) prove the effectiveness of the proposed method.

A video contains rich perceptual information, such as visual appearance, motion and audio, which can be used for understanding the activities in videos. Recent works have shown the combination of appearance (spatial) and motion (temporal) clues can significantly improve human action recognition performance in videos. To further explore the multimodal representation of video in action recognition, We propose a framework to learn a multimodal representations from video appearance, motion as well as audio data. Convolutional Neural Networks (CNN) are trained for each modality respectively. For fusing multiple features extracted with CNNs, we propose to add a fusion layer on the top of CNNs to learn a joint video representation. In fusion phase, we investigate both early fusion and late fusion with Neural Network and Support Vector Machine. Compare to existing works, (1) our work measures the benefits of taking audio information into consideration and (2) implements sophisticated fusion methods. The effectiveness of proposed approach is evaluated on UCF101 and UCF101-50 (selected subset in which each video contains audio data) for action recognition. The experimental results show that different modalities are complementary to each other and multimodal representation can be beneficial for final prediction. Furthermore, proposed fusion approach achieves 85.1% accuracy in fusing spatial-temporal on UCF101 (split 1), which is very competitive to state-of-the-art works.

In recent years, Massive Open Online Courses (MOOCs) have become a phenomenon offering the possibility to teach thousands of participants simultaneously. In the same time the platforms used to deliver these courses are still in their fledgling stages. While course content and didactics of those massive courses are the primary key factors for the success of courses, still a smart platform may increase or decrease the learners experience and his learning outcome. The paper at hand proposes the usage of an A/B testing framework that is able to be used within an micro-service architecture to validate hypotheses about how learners use the platform and to enable data-driven decisions about new features and settings. To evaluate this framework three new features (Onboarding Tour, Reminder Mails and a Pinboard Digest) have been identified based on a user survey. They have been implemented and introduced on two large MOOC platforms and their influence on the learners behavior have been measured. Finally this paper proposes a data driven decision workflow for the introduction of new features and settings on e-learning platforms.

This paper at hand describes the design and implementation of an analytics service to retrieve live usage data from students enrolled in a service-oriented MOOC platform for the purpose of learning analytics (LA) research. A real-time and extensible architecture for consolidating and processing data in versatile analytics stores is introduced.

Massive Open Online Courses (MOOCs) have revolutionized higher education by offering university-like courses for a large amount of learners via the Internet. The paper at hand takes a closer look on peer assessment as a tool for delivering individualized feedback and engaging assignments to MOOC participants. Benefits, such as scalability for MOOCs and higher order learning, and challenges, such as grading accuracy and rogue reviewers, are described. Common practices and the state-of-the-art to counteract challenges are highlighted. Based on this research, the paper at hand describes a peer assessment workflow and its implementation on the openHPI and openSAP MOOC platforms. This workflow combines the best practices of existing peer assessment tools and introduces some small but crucial improvements.

This work presents an end-to-end trainable deep bidirectional LSTM (Long-Short Term Memory) model for image captioning. Our model builds on a deep convolutional neural network (CNN) and two separate LSTM networks. It is capable of learning long term visual-language interactions by making use of history and future context information at high level semantic space. Two novel deep bidirectional variant models, in which we increase the depth of nonlinearity transition in different way, are proposed to learn hierarchical visual-language embeddings. Data augmentation techniques such as multi-crop, multi-scale and vertical mirror are proposed to prevent overfitting in training deep models. We visualize the evolution of bidirectional LSTM internal states over time and qualitatively analyze how our models "translate" image to sentence. Our proposed models are evaluated on caption generation and image-sentence retrieval tasks with three benchmark datasets: Flickr8K, Flickr30K and MSCOCO datasets. We demonstrate that bidirectional LSTM models achieve highly competitive performance to the state-of-the-art results on caption generation even without integrating additional mechanism (e.g. object detection, attention model etc.) and significantly outperform recent methods on retrieval task

We showcase a system for real-time video text recognition. The system is based on the standard workflow of text spotting system, which includes text detection and word recognition procedures. We apply deep neural networks in both procedures. In text localization stage, textual candidates are roughly captured by using a Maximally Stable Extremal Regions (MSERs) detector with high recall rate, false alarms are then eliminated by using Convolutional Neural Network (CNN ) verifier. For word recognition, we developed a skeleton based method for segmenting text region from its background, then a CNN based word recognizer is utilized for recognizing texts. Our current implementation demonstrates a real time performance for recognizing scene text by using a standard laptop with webcam. The word recognizer achieves competitive result to state-of-the-art methods by only using synthetical training data.

The number of revealed and analyzed attacks that use sophisticated and complex methods increased lately. Through the usage of such methods the attackers are able to evade existing security systems and prevent a comprehensive detection of the malicious activities. Therefore, it is often necessary to perform a manual investigation of complex attacks, to identify all steps and malicious activities that belong to an attack. One main objective of an investigation is to correlate existing events and reveal relations between different activities, to get a comprehensive overview of the attack. Due to the fact that the correlation is often done manually, this process is complex and time consuming. In this paper, we propose a new automated correlation approach that uses the tainting concept to identify related log events. The approach uses meaningful attributes as taint sources and a taint policy to propagate the taint to related events. For the identification of the correlations, it is also possible to use meta-information sources to support the correlation process. Furthermore, the tainting based approach allows to visualize the correlation results in a taint graph, which simplifies the traceability. We successfully evaluated the proposed approach with log events from a simulated attack where real world attack methods and tools were used. With the new approach it was possible to identify all events that recorded the malicious activities of the attacker and the created taint graph allowed a comprehensive retracing of the attack. Thus, the correlation approach can support investigations in an effective way, because it reduces the complexity of event correlation and the needed time.

An increasing number of network connections are encrypted to protect the confidentiality of the transferred data. Also attackers make greater use of encrypted protocols to hide from detection and to hinder investigations. Currently, most security systems (e.g., Intrusion Detection Systems (IDSs) and firewalls) cannot effectively analyze encrypted traffic. This results in "blind spots", which can put the security of a whole environment at risk. In this paper, we propose a system that is capable of investigating encrypted Remote Desktop Protocol (RDP) connections. In the first step the private RSA key of the RDP server is used to decrypt the TLS/SSL layer of the RDP stream. In the second step our system extracts all relevant information (e.g., keystrokes and transferred files) from the RDP connection. This information makes it possible to reconstruct the behavior and the activities of an attacker with a high accuracy. For the evaluation of our approach we performed a scan of 231,025 internet facing RDP systems and revealed that over 95 % of the RDP connections to these systems can be decrypted with our system.

Question and Answering (Q&A) platforms are an important source for information and a first place to go when searching for help. Q&A sites, like StackOverflow (SO), use reward systems to incentivize users to answer fast and accurately. In this paper we study and predict the response time for those questions on StackOverflow, that benefit from an additional incentive through so called bounties. Shaped by different motivations and rules these questions perform unlike regular questions. As our key finding we note that topic related factors provide a much stronger evidence than previously found factors for these questions. Finally, we compare models based on these features predicting the response time in the context of bounty questions.

To survive reboots, 802.15.4 security normally requires an 802.15.4 node to store both its anti-replay data and its frame counter in non-volatile memory. However, the only non-volatile memory on most 802.15.4 nodes is flash memory, which is energy consuming, slow, as well as prone to wear. Establishing session keys frees 802.15.4 nodes from storing anti-replay data and frame counters in non-volatile memory. For establishing pairwise session keys for use in 802.15.4 security in particular, Krentz et al. proposed the Adaptable Pairwise Key Establishment Scheme (APKES). Yet, APKES neither supports reboots nor mobile nodes. In this paper, we propose the Adaptive Key Establishment Scheme (AKES) to overcome these limitations of APKES. Above all, AKES makes 802.15.4 security survive reboots without storing data in non-volatile memory. Also, we implemented AKES for Contiki and demonstrate its memory and energy efficiency. Of independent interest, we resolve the issue that 802.15.4 security stops to work if a node's frame counter reaches its maximum value, as well as propose a technique for reducing the security-related per frame overhead.

The detection of vulnerabilities in computer systems and computer networks as well as the weakness analysis are crucial problems. The presented method tackles the problem with an automated detection. For identifying vulnerabilities the approach uses a logical representation of preconditions and postconditions of vulnerabilities. The conditional structure simulates requirements and impacts of each vulnerability. Thus an automated analytical function could detect security leaks on a target system based on this logical format. With this method it is possible to scan a system without much expertise, since the automated or computer-aided vulnerability detection does not require special knowledge about the target system. The gathered information is used to provide security advisories and enhanced diagnostics which could also detect attacks that exploit multiple vulnerabilities of the system.

Looking at recent cyber-attacks in the news, a growing complexity and sophistication of attack techniques can be observed. Many of these attacks are performed in multiple steps to reach the core of the targeted network. Existing signature detection solutions are focused on the detection of a single step of an attack, but they do not see the big picture. Furthermore, current signature languages cannot integrate valuable external threat intelligence, which would simplify the creation of complex signatures and enables the detection of malicious activities seen by other targets. We extend an existing multi-step signature language to support attack detection on normalized log events, which were collected from various applications and devices. Additionally, the extended language supports the integration of external threat intelligence and allows us to reference current threat indicators. With this approach, we can create generic signatures that stay up-to-date. Using our language, we could detect various login brute-force attempts on multiple applications with only one generic signature.

Nowadays gamification is a hot topic in the world, a lot of websites, applications and researches adapt this method to arouse users' motivation. From the past experience, gamification indeed has a positive influence on users' motivation especially in e-learning field. However, the gamification method either is hard to be applied to professional content called meaningful gamification or is negative on user's intrinsic motivation called reward-based gamification. So we study the game addiction mechanism and propose the reward-based intermittent reinforcement method in gamification to take advantage of user independence feature in the latter one and eliminate the negative influence on user's intrinsic motivation. In order to investigate the practicability and integrate effectiveness, we implement this model in our tele-teaching platform.

The blogosphere allows analysts to track opinions and sentiments of individuals, groups or the general public with large sample sizes regarding many topics. Essential for the sentiment analysis are visualizations. The visual understanding of large corpora's sentiment is far more effective than relying on textual representations of the analyzed content. Users are very interested in changes in the public opinion. Thus, the identification of patterns is of high interest. In this paper, we propose a cluster heat map visualization for sentiment visualization that displays the sentiment development of various related terms over time intervals. As we want to encourage the discovery of patterns over multiple related topics, we apply an ordering algorithm based on dimensionality reduction to the cluster heat map and improve upon the ordering algorithm to enable fast pattern recognition.

The paper at hand evaluates the Massive Open Online Course (MOOC) Spielend Programmieren Lernen (Playfully learning to program), an effort to scale the youth development program at the Hasso Plattner Institute (HPI) for a larger audience. The HPI has a strong tradition in attracting children and adolescents to take their first steps towards a career in IT at an early age. The Schülerakademie, the Schülerkolleg, the Schülerklub, and the support for the CoderDojo in Potsdam are some of the regular activities in this context to take youngsters by the hand and supply them with material and guidance in their mother tongue. With the emergence of MOOCs and the success of HPI's own MOOC Platform - openHPI - it was a natural step to develop a course to address an audience that is only marginally represented in openHPI's regular courses: school children and adolescents. A further novelty for openHPI in this course was the focus on teaching programming with a high percentage of obligatory hands-on tasks. Particularly for this course, a standalone tool allowing participants to write and evaluate code directly in the browser - without the need to install additional software - has been developed. We will compare this tool to a small selection of similar approaches on other platforms. As it will be shown, the course attracted a far more diverse audience than expected, and therefore, also needs to be seen in the context of spreading digital literacy amongst wider parts of society. In this context we also will discuss the significant differences in the usage of the forum between the course Spielend Programmieren Lernen and the course In-Memory Databases, a more traditional openHPI course.

Cross-Modal mapping plays an essential role in multimedia information retrieval systems. However, most of existing work paid much attention on learning mapping functions but neglected the exploration of high-level semantic representation of modalities. Inspired by recent success of deep learning, in this paper, deep CNN (convolutional neural networks) features and topic features are utilized as visual and textual semantic representation respectively. To investigate the highly non-linear semantic correlation between image and text, we propose a regularized deep neural network(RE-DNN) for semantic mapping across modalities. By imposing intra-modal regularization as supervised pre-training, we finally learn a joint model which captures both intra-modal and inter-modal relationships. Our approach is superior to previous work in follows: (1) it explores high-level semantic correlations, (2) it requires little prior knowledge for model training, (3) it is able to tackle modality missing problem. Extensive experiments on benchmark Wikipedia dataset show RE-DNN outperforms the state-of-the-art approaches in cross-modal retrieval.

When looking at media reports nowadays, major security breaches of big companies and governments seem to be a normal situation. An important step for the investigation or even prevention of these breaches is to normalize and analyze security-related log events from various systems in the target network. However, the number of log events produced in big IT landscapes can grow up to multiple billions per day. Current log management solutions, e.g., Security Information and Event Management (SIEM), cannot even closely normalize such huge amounts of data and therefore disable the tracking of attacks in real-time, which means that the log data remains unusable for attack analysis. In this paper, we present an approach to fully normalize event logs in high-speed by making use of established high-performance inter-thread messaging in conjunction with a hierarchical knowledge-base of log formats and parallel processing on multiple low-end systems. Using our approach, we are able to process more than 250,000 events/sec on relatively low-profile machines and can therefore easily handle more than 20 billion events/day, which is enough to handle average and peek loads of log events from big enterprise networks.

Modern Security Information and Event Management systems should be capable to store and process high amount of events or log messages in different formats and from different sources. This requirement often prevents such systems from usage of computational-heavy algorithms for security analysis. To deal with this issue, we built our system based on an in-memory data base with an integrated machine learning library, namely SAP HANA. Three approaches, i.e. (1) deep normalisation of log messages (2) storing data in the main memory and (3) running data analysis directly in the database, allow us to increase processing speed in such a way, that machine learning analysis of security events becomes possible nearly in real-time. To prove our concepts, we measured the processing speed for the developed system on the data generated using Active Directory tested and showed the efficiency of our approach for high-speed analysis of security events.

In this paper we showcase a system for real-time text detection and recognition. We apply deep features created by Convolutional Neural Networks (CNNs) for both text detection and word recognition task. For text detection we follow the common localization-verification scheme which already shown its excellent ability in numerous previous work. In text localization stage, textual regions are roughly detected by using a MSERs (Maximally Stable Extremal Regions) detector with high recall rate. False alarms are then eliminated by using a CNNs classifier, and remaining text regions are further grouped into words. In the word recognition stage, we developed an skeleton-based text binarization method for segmenting text from its background. A CNNs based recognizer is then applied for recognizing character. The initial experiments show the powerful ability of deep features for text classification comparing with commonly used visual features. Our current implementation demonstrates real-time performance for recognizing scene text by using a standard PC with webcam.

For testing new methods of network security or new algorithms of security analytics, we need the experimental environments as well as the testing data which are much as possible similar to the real-world data. Therefore, the researchers are always trying to find the best approaches and recommendations of creating and simulating testbeds, because the issue of automation of the testbed creation is a crucial goal to accelerate research progress. One of the ways to generate data is simulate the user behavior on the virtual machines, but the challenge is how to describe what we want to simulate.

The detection of vulnerabilities in computer systems and computer networks as well as the representation of the results are crucial problems. The presented method tackles the problem with an automated detection and an intuitive representation. For detecting vulnerabilities the approach uses a logical representation of preconditions and postconditions of vulnerabilities. Thus an automated analytical function could detect security leaks on a target system. The gathered information is used to provide security advisories and enhanced diagnostics for the system. Additionally the conditional structure allows us to create attack graphs to visualize the network structure and the integrated vulnerability information. Finally we propose methods to resolve the identified weaknesses whether to remove or update vulnerable applications and secure the target system. This advisories are created automatically and provide possible solutions for the security risks.

User-centered designers often seek to synthesize data from user research into insights and a shared point of view among team members. This paper explores the synthesis process and opportunities for providing computational support. First, we present interviews with novice and expert designers on the common practices and challenges of syn-thesis. Based on these interviews, we developed digital whiteboard software support for sorting individual seg-ments of user research. The system separates out individual and group activity and helps the team externalize and syn-thesize their different views of the data. Through a case study, we explore two computer-supported approaches: a structured condition that externalizes the different perspec-tives on the data of each team member and an unstructured condition that allows each member to organize data into clusters. Novice designers tended to prefer the structured synthesis process, while more experienced designers pre-ferred to freely arrange information segments and create clusters on their own. We provide implications for design education and support tools for user research synthesis.

Being able to identify locations associated to a Web resource is essential for providing location-based Web applications. However, geographical information in Web documents is rarely supplied in a machine-readable way and therefore not easily discoverable. As a consequence, it is necessary to extract geographical keywords from Web documents and to associate locations with them. This method is called location tagging. In this paper we present a location tagging approach for unstructured documents which utilizes multiple external location providers. Detected locations are ranked according to their relevance for the document, in order to identify a document's geographical focus, which is its most representative location. We present an exemplary implementation of our proposed approach using two location providers and evaluate our method's applicability.

Information about upcoming trends is considered to be a valuable source of knowledge for both, companies and individuals. A large number of market analysts working at monitoring a particular business field, with many employing manual methods to do so. Since the amount of available data on the internet is far too high for humans to monitor, which carries a major risk of substantial amount of information being missed, the necessity arose to detect emerging trends automatically. Weblogs are an important medium to publish information and discuss certain topics. The web platform BlogIntelligence analyzes and visualizes the content and interconnection of blogs in the blogosphere. One area of focus is the detection of trends over a period of time, which is especially helpful for product vendors. But even more interesting, views expressed in weblog posts influence the reader's opinion. Integrating the strength and direction of expressed sentiments enhances the trend detection significantly. In this work, we introduce an approach to enrich the trend detection with sentiment analysis.

A lot of research efforts are going on in the area of mining emotions within the world wide web. The BlogIntelligence application is analyzing tons of blog posts and extracts emotions out of this big amount of data. Therefore we thought about how to visualize these emotions in a very meaningful way. While we applied a smart map as a proven technique, we overcame conceptual and technical challenges to provide a feasible utility.

The process of assessment grading does not end on publishing the grades and assessment results. Often, discussions about single grades arise in post-exam reviews. While in traditional teaching and smaller online classes this can be handled on a per-user based approach, in Massive Open Online Courses (MOOCs) with thousands of e-assessment submissions this approach cannot be followed without investing huge resources. Therefore, a new approach of handling re-gradings is needed. In this paper, we will discuss the experience on re-gradings based on six courses offered on openHPI. We will introduce our concept of automated re-grading following an approach of fairness and demonstrate the importance of transparent communication. This paper provides a blueprint for handling re-grading issues in large scale learning environments.

There is a great demand for hands-on training in engineering education. In the context of a Massive Open Online Course (MOOC), assessing these experiments manually by teaching assistants is not possible owed to the high number of participants and the resulting workload for the teaching team. Systems for machine-based assessment of coding tasks are existing, but not necessarily available publicly, or not prepared to handle the massive amount of users in a MOOC. Definitely, they are not available “ad hoc”, but require a certain amount of effort to be integrated in the MOOC platform or to be made available for the students in another way. Time and money to provide the required effort is not always available. This work presents a lightweight solution for the assessment of practical programming exercises, based on third party online coding tools. The solution was introduced as a part of openHPI's Web-Technologies course. The basic idea is to prepare a task in an available online tool, along with a piece of code that is able to evaluate the participant's solution. In case of success the participant is provided with a password, which in return serves as the answer for a fill-in-the-gap question in a standard quiz as provided by the openHPI MOOC platform, and thus allows for automatic online assessment based on practical coding exercises.

Public cloud storage services enable organizations to manage data with low operational expenses. However, the benefits come along with challenges and open issues such as security, reliability and the risk to become dependent on a provider for its service. In our previous work, we presented a system that improves availability, confidentiality and reliability of data stored in the cloud. To achieve this objective, we encrypt user's data and make use of the RAID-technology principle to manage data distribution across cloud storage providers. Recently, we conducted a proof-of-concept experiment for our application to evaluate the performance and cost effectiveness of our approach. We observed that our implementation improved the perceived availability and, in most cases, the overall performance when compared with cloud providers individually. We also observed a general trend that cloud storage providers have constant throughput values - whereby the individual throughput performance differs strongly from one provider to another. With this, the experienced transmissions can be utilized to increase the throughput performance of the upcoming data transfers. The aim is to distribute the data across providers according to their capabilities utilizing the maximum of the available throughput capacity. To assess the feasibility of the approach we have to understand how providers handle high simultaneous data transfers. Thus, in this paper we focus on the performance and the scalability evaluation of particular cloud storage providers. To this end, we deployed our application using eight commercial cloud storage repositories in different countries and conducted a set of extensive experiments.

The differences in log file formats employed in a variety of services and applications remain to be a problem for security analysts and developers of intrusion detection systems. The proposed solution, i.e. the usage of common log formats, has a limited utilization within existing solutions for security management. In our paper, we reveal the reasons for this limitation. We show disadvantages of existing common log formats for normalisation of security events. To deal with it we have created a new log format that fits for intrusion detection purposes and can be extended easily. Taking previous work into account, we would like to propose a new format as an extension to existing common log formats, rather than a standalone specification.

Cloud computing has emerged over the last years as a new model of delivering computation to consumers. Due to the nature of the cloud, where the data and the computation are beyond the control of the user, data privacy and security becomes a vital factor in this new paradigm. Several research studies reported that security and privacy is cited as the biggest concern in adopting cloud computing. In this paper, we present HPISecure, a software prototype designed to facilitate the process of securing data stored on the cloud. HPISecure allows the user to store an encrypted version of his data on the cloud without breaking the functionality of the application. HPISecure intercepts the HTTP request/response objects, encrypt data before transmitting to the cloud, and decrypt data received back from the server. We have successfully tested HPISecure with Google Docs and Google Calender. Future work includes extending HPISecure to work with enterprise applications.

Creative ways of working with whiteboards and sticky notes are growing in popularity even in global companies. However, digital tools for enabling these ways of working, especially for geographically distributed teams, have still not been adopted in these companies. We present Tele-Board, a web-based digital whiteboard and sticky note system and describe how it was used in a large company at three locations. From system log data and interviews recorded after three months of use, we found that idea generation and feedback collection can be facilitated if a system offers real-time synchronous editing as well as asynchronous input. Interestingly, the users who were not located at the company's headquarters regarded the tool as very beneficial and used it more than their colleagues at the headquarters. We provide a detailed analysis of the study and important points for fostering the adoption of creative tools in large companies.

Mobile devices such as smartphones and tablets play an increasing role in today's working environment. The variety of computer platforms increased in the same way, which makes the development of cross-platform applications even more challenging. Tele-Board is a real-time remote collaboration system based on the Java programming language. Therefore, it cannot be run on most mobile devices. In order to overcome this limitation, we redeveloped the system on the basis of HTML5 technology. We present an approach for combining web based networking and rendering in a single application for real-time collaboration based on SVG, HTML5 Canvas, Websockets, and Web workers. In our prototype we implemented optimization mechanisms leveraging the Canvas API's rendering flexibility. This way, our canvas based rendering performs better than a respective SVG version. Moreover, our solution integrates server communication effectively so that the rendering performance is hardly influenced by user input.

Such information as system and application logs as well as the output from the deployed security measures, e.g., IDS alerts, firewall logs, scanning reports, etc., is important for the administrators or security operators to be aware at first time of the running state of the system and take efforts if necessary. In this context, high performance security analytics is proposed to address the challenges to rapidly gather, manage, process, and analyze the large amount of real-time information generated from the large scale of enterprise IT-Infrastructure while it is being operated. As an example of next generation Security Information and Event Management (SIEM) platform, Security Analytics Lab (SAL) has been designed and implemented based on the newly emerged In-Memory data management technique, which makes it possible to efficiently organize and access different types of event information through a consistent central storage and interface. To correlate the information from different sources and identify the meaningful information is another challenging task, which makes great sense for quickly judging the current situation and making the decision. In this paper, the multi-core processing technique is introduced in the SAL platform. Various correlation algorithms, e.g., k-means based algorithms, ROCK and QROCK clustering algorithms, have been implemented and integrated in the multi-core supported SAL architecture. Practical experiments are conducted and analyzed to proof that the performance of analytics can be significantly improved by applying multi-core processing technique in SAL.

Looking at current IDS and SIEM systems, we observe heavy processing power dedicated solely to answering a simple question, What is the format of the log line that the IDS (or SIEM) system should process next? Due to the apparent difficulties of uniquely identifying a log line at run-time, most systems today do little or no normalisation of the events they receive. Indeed these systems often rely on popular search engine applications for processing and analysing the event information they receive, which results in slower and far less accurate event correlations. In this process, a large list of tokenisers is usually created in order to find an answer to the above posted question. The tokenisers are run against the log lines, until a match is found. The appropriate log line can then be passed on to the correct extraction module for further processing. This process is currently the standard procedure of most IDS and SIEM systems. To address this problem and to optimise and improve the said process, this paper describes a method for detecting the exact type and format of a read log line in the first place. The method presented performs in an efficient manner, while it is less resource hungry. The proposed detection system is described and implemented, its pros and cons are analysed and weighed against methods currently implemented by popular IDS and SIEM systems for solving this task.

Recently a new format of online education has emerged that combines video lectures, interactive quizzes and social learning into an event that aspires to attract a massive number of participants. This format, referred to as Massive Open Online Course (MOOC), has garnered considerable public attention, and has been invested with great hopes (and fears) of transforming higher education by opening up the walls of closed institutions to a world-wide audience. In this paper, we present two MOOCs that were hosted at the same platform, and have implemented the same learning design. Due to their difference in language, topic domain and difficulty, the communities that they brought into existence were very different. We start by describing the MOOC format in more detail, and the distinguishing features of openHPI. We then discuss the literature on communities of practice and cultures of participation. After some statistical data about the first openHPI course, we present our qualitative observations about both courses, and conclude by giving an outlook on an ongoing comparative analysis of the two courses.

Cloud Computing as a service-on-demand architecture has grown in importance over the previous few years. One driver of its growth is the ever increasing amount of data which is supposed to outpace the growth of storage capacity. The usage of cloud technology enables organizations to manage their data with low operational expenses. However, the benefits of cloud computing come along with challenges and open issues such as security, reliability and the risk to become dependent on a provider for its service. In general, a switch of a storage provider is associated with high costs of adapting new APIs and additional charges for inbound and outbound bandwidth and requests. In this paper, we present a system that improves availability, confidentiality and reliability of data stored in the cloud. To achieve this objective, we encrypt user's data and make use of the RAID-technology principle to manage data distribution across cloud storage providers. We conduct a proof-of-concept testbed experiment for our application to evaluate the performance and cost effectiveness of our approach. We deployed our application using eight commercial cloud storage repositories in different countries. Our approach allows users to avoid vendor lock-in, and reduces significantly the cost of switching providers. We also observed that our implementation improved the perceived availability and, in most cases, the overall performance when compared with individual cloud providers. Moreover, we estimated the monetary costs to be competitive to the cost of using a single cloud provider.

In this paper we propose a solution which segments lecture video by analyzing its supplementary synchronized slides. The slides content derives automatically from OCR (Optical Character Recognition) process with an approximate accuracy of 90%. Then we partition the slides into different subtopics by examining their logical relevance. Since the slides are synchronized with the video stream, the subtopics of the slides indicate exactly the segments of the video. Our evaluation reveals that the average length of segments for each lecture is ranged from 5 to 15 minutes, and 45% segments achieved from test datasets are logically reasonable.

Internet scalability depends on scalability of its core routing protocol - Border Gateway Protocol (BGP). However, dynamics of BGP still conceal many unanswered questions. Most of these questions are related to BGP update messages: root cause of update spikes, correlation between update spikes in the different parts of the Internet and influence of individual spikes on global routing. This article presents a methodology to locate routing events behind specific BGP update spikes. The method explores correlated updates seen on different vantage points [1]. Although previous work [2] uses similar approach to identify origin of update bursts, we revise the question considering one-second update spikes as a point of view. This concept allows not only to identify an area where the routing event has happened, but also to find, how an individual BGP update spike was formed, i.e. find a propagation path for a set of routing events behind the spike. Revealed propagation paths - if analysed for a significant amount of update messages - could tell us new facts about specific types of routing events and improve our understanding of BGP scalability.

The Domain Name System (DNS) is an essential part of the Internet on whose function many other protocols rely. One key DNS function is Dynamic Update, which allows hosts on the network to make updates to DNS records dynamically, without the need for restarting the DNS service. Unfortunately, this dynamic process does expose DNS servers to security issues. To address these issues two protocols were introduced: Transaction Signature (TSIG) and Domain Name System Security Extensions (DNSSEC). In Internet Protocol version 4 (IPv4) networks using these protocols eliminated security issues. In Internet Protocol version 6 (IPv6) however, there is an issue with the DNS authentication process when using the Stateless Address Auto Configuration (SLAAC) mechanism (new to IPv6, nonexistent in IPv4). This authentication issue occurs when a node wants to update its resource records on a DNS server, during the DNS update process, or when a client wants to authenticate a DNS resolver to ensure that the DNS response does not contain a spoofed source address or message. In this paper we propose the use of a new mechanism which makes use of asymmetric cryptography to establish a trust relationship with the DNS server. We also consider the use of the current security parameters used to generate IPv6 addresses in a secure manner, i.e. Secure Neighbor Discovery (SeND), for assuring clients and DNS servers that the one they are communicating with is the real owner of this IP address. Since we are extending the RDATA field within the TSIG protocol to accommodate these new security parameters, we will call this new mechanism the CGA-TSIG algorithm.

The default method for IPv6 address generation uses an Organizationally Unique Identifier (OUI) assigned by the IEEE Standards Association and an Extension Identifier assigned by the hardware manufacturer (RFC 4291). For this reason a node will always have the same Interface ID (IID) whenever it connects to a new network. Because the node's IP address does not change, the node will be vulnerable to privacy related attacks. Currently this problem is addressed by the use of two mechanisms that do not use MAC addresses or other unique values for randomizing the IID during its generation: Cryptographically Generated Addresses (CGA) (RFC 3972) and Privacy Extension (RFC 4941). The problem with the former approach is the computational cost involved in the IID generation and, more importantly, the verification process. The problem with the latter approach is the lack of necessary security mechanisms and that it provides the node with only partial protection against privacy related attacks. This document proposes the use of a new algorithm in the generation of the IID to reduce computational cost while, at the same time, securing the node against some types of attack, like IP spoofing. These attacks are prevented by the addition of a signature to messages sent over the network and by direct use of a public key in the IP address.

In IPv6 networks, two security mechanisms are available at the network-layer; SEcure Neighbor Discovery (SEND) and IP security (IPsec). Although both provide authentication, neither subsumes the other; both SEND and IPsec mechanisms should be deployed together to protect IPv6 networks. However, when a node uses both SEND and IPsec, the authentication has to be done twice, which increases the burden on the node and decreases its performance. In this paper, we propose an approach to enable them to work together under the mediation of an Authentication Management Block, where IPsec uses the public-private keys obtained by SEND rather than negotiating its own authentication credentials in order to save the time and facilitate the IPsec authentication deployment. We implement and evaluate our approach using ipsec-tools and DoCoMo SEND implementations. Our proof-of-concept experiment shows a considerable speedup of IPsec authentication time.

Privacy is a very important element in every one's everyday life. Most users would not like to have their data exposed to other people on the Internet. The initial approach used for attacking a user's privacy and security is done by scanning the nodes on a network. This gives an attacker the ability to obtain the IP addresses in use by this node so that this information can then be used to initiate further attacks against this node, such as tracking them via their IP address across the networks, and then, later correlating the user's activities with his IP address. The first attempt by the Internet Engineering Task Force (IETF) to protect a user's privacy was defined in the Privacy Extension RFC [13]. Unfortunately this RFC has some deficiencies which makes its use vulnerable to privacy related attacks. To address this problem, and solve the deficiencies that exist with the use of this RFC, we introduce our new algorithm, which not only maintains a node's lifetime, but also provides a user with a method for randomized Interface ID (IID) generations.

Security has recently become a very important concern for entities using IPv6 networks. This is especially true with the recent news reports where governments and companies have admitted to credible cyber attacks against them in which confidential information and the security of data have been compromised. In this paper we will introduce a flexible framework that can be used for penetration testing of IPv6 networks. Due to the large address space in each of the IPv6 subnets, the traditional scanning approaches do not work. Here we introduce our new scanning algorithm which will find the IPv6 nodes on the Internet which are using Domain Name System (DNS) servers. Our implementation results showed that the use of the DNS Security Extension (DNSSEC) with NSEC3 [4], which is a new and promising approach for the prevention of zone walking, was not able to prevent us from gathering information about nodes on different networks.

Current ranking algorithms, such as Page Rank, Technorati authority, and BI-Impact, favor blogs that report on a diversity of topics since those attract a large audience and thus more visitors, links, and comments. On the other side, niche blogs with a very specific topic only attract a small audience and thus have only a small reach. This results in a low ranking from today's blog retrieval systems. We argue that the consistency of a blog, i.e. how focused an author reports on a single topic, is a sign for expert knowledge. To find these blogs is particular important for other domain experts to identify blogs that they would like to follow and stay in active contact. To ease the retrieval of expert blogs, i.e. to separate them from the mass of blogs that report on random topics, we introduce a metric for blogs based on topic consistency. We divide the consistency ranking in four different aspects: (1) intra-post, (2) inter-post, (3) intra-blog, and (4) inter-blog consistency. By evaluating the metric with a test data set of 12,000 crawled blogs, we demonstrate the plausibility of our approach.

Energy management has become a significant concern in data centers to reduce operational costs and maintain systems' reliability. Using virtualization allows server consolidation, which increases server utilization and reduces energy consumption by turning off unused servers. However, server consolidation and turning off servers can cause also consequences if they are not exploited efficiently. For instance, many researchers consider a deterministic demand for capacity planning, but the demand is always subject to uncertainty. This uncertainty is an outcome of the workload prediction and the workload fluctuation. This paper presents a robust optimization for proactive capacity planning. We do not presume that the demand of VMs is deterministic. Thus, we implement a range prediction approach instead of a single point prediction. Then, we implement a robust optimization model exploiting the range-based prediction to determine the number of active servers for each capacity planning period. The results of the simulation show that our approach can mitigate undesirable changes in the power-state of the servers. Additionally, the results indicate an increase in the servers' availability for hosting new VMs and reliability against a system failure during power-state changes. As future work, we intend to apply our approach to dynamic workload such as a web application. We plan to investigate applying our approach to other resources, where we consider only the CPU demand of VMs. Finally, we compare our approach against the approaches using stochastic optimization.

Traditionally, Infrastructure as a Service (IaaS)providers deliver their services as Reserved or On-Demand instances. Spot Instances (SIs) is a complementary service that allows customers to bid on the free capacity at the provider data centers. Therefore, the decrease in the free capacity may result in terminating instances abruptly. To ensure fair trading, the provider does not charge customers for the interrupted partial hours. However, SIs price history traces analysis shows that uncharged time could rise up to 30% of the instance total run time, which means a reduction in the provider's profit. In this paper, we propose Elastic Spot Instances (ESIs) approach. It is a trade-off between the price and the total run time, where instead of abruptly terminating the SIs, the provider scales down their capacity proportionally to the increase in the price. Our approach delegates the task of interrupting the instances into the customers, but at the same time keeps the control on the provider side to isolate SIs' impact on the other services at overloaded time. Our approach doesn't imply an additional overhead or complex modification to current IaaS, while it consumes interfaces that are available by most of nowadays virtualization technologies.

Recently, the advance of cloud computing services has attracted many customers to host their Internet applications in the cloud. Infrastructure as a Service (IaaS) is on top of these services where it gives more control over the provisioned resources. The control is based on online monitoring of specific metrics (e.g., CPU, Memory, and Network). Despite the fact that these metrics guide resources provisioning, the lack of understanding application behavior can lead to wrong decisions. Moreover, current monitored metrics alone do not help in resources contention prediction, which is very common in shared infrastructures like IaaS. Nevertheless, the architecture of Internet applications, as multi-tier systems, makes contention prediction more complex while its influence can migrate from one tier to another. In this paper, we propose a pro-active global controller not only for dynamic resources provisioning, but also for predicting and eliminating contentions in multi-tier applications. Our technique combines monitored metrics, which are provided by current IaaS providers, with models that are built depending on the Internet applications profiling. The fitness of the monitored metrics to the application model is used for contention prediction. We examined our technique using RUBiS benchmark. The results express the efficiency of the developed algorithms in maintaining Internet applications performance even in shared infrastructures.

Reputation systems aim to provide a mechanism for establishing trust for online interactions attempting to mimic their real-world counterparts. Reputation-based approaches depend on the user’s local experiences and feedback to create a soft measure for trust decision. They use various clues and past

Online (self) assessment is an important functionality e-learning courseware, especially if the system is intended for use in distant learning courses. Precisely for hands-on exercises, the implementation of effective and cheating-proof assessment tests poses a great challenge. That is because of the static characteristics of exercise scenarios in the laboratories: adopting the environment for the provision of a “unique” hands-on experience for every student in a manual manner is connected with enormous maintenance efforts and thus not scalable to a large number of students. This work presents a software solution for the assessment of practical exercises in an online lab based on virtual machine technology. The basic idea is to formally parameterize the exercise scenarios and implement a toolkit for the dynamic reconfiguration of virtual machines in order to adopt the defined parameters for the training environment. The actual values of these parameters come to use again in the dynamic generation of multiple-choice or free-text answer tests for a web-based e-assessment environment.

SUMMARY Intrusion Detection Systems (IDS) have been widely deployed in practice for detecting malicious behavior on network communication and hosts. False‐positive alerts are a popular problem for most IDS approaches. The solution to address this problem is to enhance the detection process by correlation and clustering of alerts. To meet the practical requirements, this process needs to be finished fast, which is a challenging task as the amount of alerts in large‐scale IDS deployments is significantly high. We identifytextitdata storage and processing algorithms to be the most important factors influencing the performance of clustering and correlation. We propose and implement a highly efficient alert correlation platform. For storage, a column‐based database, an In‐Memory alert storage, and memory‐based index tables lead to significant improvements of the performance. For processing, algorithms are designed and implemented which are optimized for In‐Memory databases, e.g. an attack graph‐based correlation algorithm. The platform can be distributed over multiple processing units to share memory and processing power. A standardized interface is designed to provide a unified view of result reports for end users. The efficiency of the platform is tested by practical experiments with several alert storage approaches, multiple algorithms, as well as a local and a distributed deployment. Copyright © 2011 John Wiley & Sons, Ltd.

We present the digital whiteboard system Tele-Board, which automatically captures all interactions made on the all-digital whiteboard and thus offers possibilities for a fast interpretation of usage characteristics. Analyzing team work at whiteboards is a time-consuming and error-prone process if manual interpretation techniques are applied. In a case study, we demonstrate how to conduct and analyze whiteboard experiments with the help of our system. The study investigates the role of video compared to an audio-only connection for distributed work settings. With the simplified analysis of communication data, we can prove that the video teams were more active than the audio teams and the distribution of whiteboard interaction between team members was more balanced. This way, an automatic analysis can not only support manual observations and codings, but also give insights that cannot be achieved with other systems. Beyond the overall view on one sessions focusing on key figures, it is also possible to find out more about the internal structure of a session.

This paper presents the state of the art in the development of Semantic-Web-enabled software using object-oriented programming languages. Object triple mapping (OTM) is a frequently used method to simplify the development of such software. A case study that is based on interviews with developers of OTM frameworks is presented at the core of this paper. Following the results of the case study, the formalization of OTM is kept separate from optional but desirable extensions of OTM with regard to metadata, schema matching, and integration into the Semantic-Web infrastructure. The material that is presented is expected to not only explain the development of Semantic-Web software by the usage of OTM, but also explain what properties of Semantic-Web software made developers come up with OTM. Understanding the latter will be essential to get nonexpert software developers to use Semantic-Web technologies in their software.

For several years online reputation systems have been evolving in an attempt to mimic their real-world counterparts. Rating in online communities comes in many forms such as numbers, stars, scales, etc.. The underlying aggregation or computation methods varies according to the community's informatio

We present the digital whiteboard system Tele-Board, which automatically captures all interactions made on the all-digital whiteboard and thus offers possibilities for a fast interpretation of usage characteristics. Analyzing team work at whiteboards is a time-consuming and error-prone process if ma

Social networking functions make their way through more and more applications for private and professional use as they encourage participation and content contribution. However, designing and implementing components that attract users and truly support content creation is not a trivial task. In this

Remote collaboration among geographically dispersed team members has become standard practice for many companies and research teams. A number of computer supported collaborative work systems exist, but there still lacks acceptable support for teams working in creative settings, where traditionally numerous physical and analog tools are used. We have created an environment for teams applying creative methods that allows them to work together efficiently across distances, without having to change their working modes. We present the Tele-Board system, which combines video conferencing with a synchronous transparent whiteboard overlay. This setup enables regionally separated team members to simultaneously manipulate artifacts while seeing each other's gestures and facial expressions. Our system's flexible architecture maximizes hardware independence by supporting a diverse selection of input devices.

Power management is one of the biggest challenges facing current data centers. As processors consume the dominant amount of power in computer systems, power management of multicore processors is extremely significant. An efficient power model that accurately predict the power consumption of a processor is required to develop effective power management techniques. However, this challenge rises with using virtualization and increasing number of cores in the processors. In this paper, we analyze power consumption of a multicore processor, we develop three statistical CPU-Power models based on the number of active cores and average running frequency using a multiple liner regression. Our models are built upon a virtualized server. The models are validated statistically and experimentally. Statistically, our models cover 97\% of system variations. Furthermore, we test our models with different workloads and three benchmarks. The results show that our models achieve better performance compared to the recently proposed model for power management in virtualized environments. Our models provide highly accurate predictions for un-sampled combinations of frequency and cores, 95\% of the predicted values have less than 7\% error. Thus, we can integrate these models into power management mechanisms for a dynamic configuration of a virtual machine in terms of the number of its virtual-CPUs and the frequency of physical cores to achieve both performance and power constrains.

Identity federation denotes a concept for the controlled sharing of user authentication and user attributes between independent trust domains. Using WS-Federation, service providers and identity providers can set up a Circle of Trust, a so called federation, in which each member is willing to trust on assertions made by another partner. However, if a member has to rely on information received from a foreign source, the need for assurance that the information is correct is a natural requirement prior to using it. Identity assurance frameworks exist that can be used to assess the trustworthiness of identity providers. The result of this assessment is a level of trust, that can be assigned to an identity provider. However, existing approaches for evaluating identity assurance do not allow to define trust levels for individual attributes. In our trust model, we consider both: (a) trust in an identity provider as the issuer of assertions and (b) trust in single attributes that an identity provider manages. In this paper, we show how our approach that we implemented in a logic-based framework can be used in web service scenarios to provide trust information on the level of identity attributes, especially about the verification process, and to match trust requirements of attributes during request processing.

Today, there is a huge amount of security services that can be used to implement different security requirements in Web Service based systems. For example, identity management services are required for authentication and authorization whereas message logging services are necessary to achieve non-repudiation. However, the deployment and configuration of these security services usually requires expert knowledge about the systems and expert knowledge about security requirements and implementations which a person can only learn by experience. Furthermore, today's Web Service based systems become increasingly complex. Thus, implementing security requirements is a complex and error prone task, even for experts. For this paper, we analysed several service-based implementations for identity management and their differences in the service orchestration. We present an approach to derive the needed security services, their configuration, and their connections to the functional services, based on defined security requirements for a Web Service based system. Therefore, we evaluate the UML use case model of the system and apply service security pattern derived during the analysis of the identity management implementations.

Modern attacks are using sophisticated and innovative techniques. The utilization of cryptography, self-modified code, and integrated attack frameworks provide more possibilities to circumvent most existing perimeter security approaches, such as firewalls and IDS. Even Application Layer Gateways (ALG) which enforce the most restrictive network access can be exploited by using advanced attack techniques. In this paper, we propose a new attack for circumventing ALGs. By using polymorphic and encrypted shellcode, multiple shellcode stages, protocol compliant and encrypted shell tunneling, and reverse channel discovery techniques, we are able to effectively bypass ALGs. The proposed attack consists of four phases with certain requirements and results. We implemented the initial shellcode as well as the different stages and conducted the practical attack using an existing ALG. The possibility to prevent this attack with existing approaches is discussed and further research in the area of perimeter security and log management is motivated.

During the last years, digital lecture libraries and lecture video portals have become more and more popular. However, finding efficient methods for indexing multimedia still remains a challenging task. Since the text displayed in a lecture video is closely related to the lecture content, it provides a valuable source for indexing and retrieving lecture contents. In this paper, we present an approach for automatic lecture video indexing based on video OCR technology. We have developed a novel video segmenter for automated slide video structure analysis and a weighted DCT (discrete cosines transformation) based text detector. A dynamic image constrast/brightness adaption serves the purpose of enhancing the text image quality to make it processible by existing common OCR software. Time-based text occurence information as well as the analyzed text content are further used for indexing. We prove the accuracy of the proposed approach by evaluation.

Neighbor Discovery Protocol (NDP) is an essential protocol in IPv6 suite, but it is known to be vulnerable to critical attacks. Thus, SEcure Neighbor Discovery (SEND) is proposed to counter NDP security threats. Unfortunately, operating systems lack the sophisticated implementations for SEND. There is limited success with SEND implementation for Linux and BSD, and no implementation for Windows families. Therefore, the majority of the users are not secured with SEND. In this paper, we will introduce an implementation of SEND for Windows families (WinSEND). WinSEND is a user-space application which provides the protection for NDP in Windows. It has direct access to Network Interface Card (NIC) and efficiently handles NDP messages by using Winpcap. WinSEND works as a service with easy user interface to set the security parameters for selected NIC.

Reputation has been explored in diverse disciplines such as artificial intelligence, electronic commerce, peer-to-peer network, and multi-agent systems. Recently it has been a vital component for ensuring trust in web services and service oriented architecture domains. In this paper, we show details about our context-aware reputation framework. The framework is based on our semantic representation model for reputation called Reputation Object (RO) model. We discuss the advantages and propositions to construct such framework, its components, and how it is implemented. The importance of developing and using such generic reputation framework is highlighted within the emergence of the Semantic Web and service oriented architecture.

In this article, we discuss the notions of experts and expertise in resource discovery in the context of collaborative tagging systems. We propose that the level of expertise of a user with respect to a particular topic is mainly determined by two factors. First, an expert should possess a high‐quality collection of resources, while the quality of a Web resource in turn depends on the expertise of the users who have assigned tags to it, forming a mutual reinforcement relationship. Second, an expert should be one who tends to identify interesting or useful resources before other users discover them, thus bringing these resources to the attention of the community of users. We propose a graph‐based algorithm, SPEAR (spamming‐resistant expertise analysis and ranking), which implements the above ideas for ranking users in a folksonomy. Our experiments show that our assumptions on expertise in resource discovery, and SPEAR as an implementation of these ideas, allow us to promote experts and demote spammers at the same time, with performance significantly better than the original hypertext‐induced topic search algorithm and simple statistical measures currently used in most collaborative tagging systems.

Intrusion Detection Systems (IDS) have been used widely to detect malicious behavior in network communication and hosts. IDS management is an important capability for distributed IDS solutions, which makes it possible to integrate and handle different types of sensors or collect and synthesize alerts generated from multiple hosts located in the distributed environment. Sophisticated attacks are difficult to detect and make it necessary to integrate multiple data sources for detection and correlation. Attack graph (AG) is used as an effective method to model, analyze, and evaluate the security of complicated computer systems or networks. The attack graph workflow consists of three parts: information gathering, attack graph construction, and visualization. This paper proposes the integration of the AG workflow with an IDS management system to improve alert and correlation quality. The vulnerability and system information is used to prioritize and tag the incoming IDS alerts. The AG is used during the correlation process to filter and optimize correlation results. A prototype is implemented using automatic vulnerability extraction and AG creation based on unified data models.

The massive adoption of social media has provided new ways for individuals to express their opinions online. The blogosphere, an inherent part of this trend, contains a vast array of information about a variety of topics. It is thus a huge think tank that creates an enormous and ever-changing archive of open source intelligence. Modeling and mining this vast pool of data to extract, exploit and describe meaningful knowledge in order to leverage (content-related) structures and dynamics of emerging networks within the blogosphere is the higher-level aim of the research presented here. This paper focuses on this project's initial phase, in which the above-mentioned data of interest needs to be collected and made available offline for further analyses. Our proprietary development of a tailor-made feed-crawler meets exactly this need. The main concept, the techniques and the implementation details of the crawler thus form the main interest of this paper and furthermore provide the basis for future project phases.

Simulation is an efficient way to model a real system by a computer program and is used to study and evaluate the characteristics or behaviors of the system. In this paper, we present an effective simulation tool that is designed for simulating the working procedure of the Lock-Keeper system, which is a high level security device for physically separating two networks. Due to the special mechanism of data exchange within the Lock-Keeper system, it is a challenging task to specify the Lock-Keeper's performance for a given application scenario. To get enough performance data, the intuitive way is to practically conduct numerous testing experiments and then analyze their results, which is extremely time consuming. Therefore, we are motivated to design and implement a simulator to predict the Lock-Keeper performance theoretically. Compare with most of available network simulation tools, the proposed simulator is capable of simulating the transfer of application layer data, i.e., file based data streams. The simulator is built based on a simple model of the Lock-Keeper data exchange procedure. Information on a target application scenario can be specified within an XML file, which is used as the input for the later calculation. Several kinds of reports are generated by this specialized simulator to indicate how the data is exchanged through Lock-Keeper. To verify the simulation results, we conduct several experiments, which practically test data transfer for the scenarios using the real Lock-Keeper system. The comparison between theoretical simulation and practical testing proves the effectiveness of our proposed simulation tool.

Remote Collaboration in synchronous and in asynchronous communication settings demands for highly specialized solutions. The requirements are set even higher when teams are working creatively with methodologies such as Design Thinking where challenging and unusual problems are addressed. Based on ou

Service-oriented Architectures support the provision, discovery, and usage of services in different application contexts. The Web Service specifications provide a technical foundation to implement this paradigm. Moreover, mechanisms are provided to face the new security challenges raised by SOA. To enable the seamless usage of services, security requirements can be expressed as security policies (e.g. WS-Policy and WS-SecurityPolicy) that enable the negotiation of these requirements between clients and services. However, the codification of security policies is a difficult and error-prone task due to the complexity of the Web Service specifications. In this paper, we introduce our model-driven approach that facilitates the transformation of architecture models annotated with simple security intentions to security policies. This transformation is driven by security configuration patterns that provide expert knowledge on Web Service security. Therefore, we will introduce a formalised pattern structure and a domain-specific language to specify these patterns.

Intrusion Detection Systems (IDS) have been widely deployed in practice for detecting malicious behavior on network communication and hosts. The problem of false-positive alerts is a popular existing problem for most of IDS approaches. The solution to address this problem is correlation and clustering of alerts. To meet the practical requirements, this process needs to be finished as soon as possible, which is a challenging task as the amount of alerts produced in large scale deployments of distributed IDS is significantly high. We identify the data storage and processing algorithms to be the most important factors influencing the performance of clustering and correlation. We propose and implement the utilization of memory-supported algorithms and a column-oriented database for correlation and clustering in an extensible IDS correlation platform. The utilization of the column-oriented database, an In-Memory Alert Storage, and memory-based index tables leads to significant improvements on the performance. Different types of correlation modules can be integrated and compared on this platform. A plugin concept for Receivers provides flexible integration of various sensors and additional IDS management systems. The platform can be distributed over multiple processing units to share memory and processing power. A standardized interface is designed to provide a unified view of result reports for end users. The efficiency of the proposed platform is tested by practical experiments with several alert storage approaches, different simple algorithms, as well as local and distributed deployment.

Cloud computing enables the provisioning of dynamically scalable resources as a service. Next to cloud computing, the paradigm of Service-oriented Architectures emerged to facilitate the provisioning of functionality as services. While both concepts are complementary, their combination enables the flexible provisioning and consumption of independently scalable services. These approaches come along with new security risks that require the usage of identity and access management solutions and information protection. The requirements concerning security mechanisms, protocols and options are stated in security policies that configure the interaction between services and clients in a system. In this paper, we present our cloud-based Service Security Lab that supports the on-demand creation and orchestration of composed applications and services. Our cloud platform enables the testing, monitoring and analysis of Web Services regarding different security configurations, concepts and infrastructure components. Since security policies are hard to understand and even harder to codify, we foster a model-driven approach to simplify the creation of security configurations. Our model-driven approach enables the definition of security requirements at the modelling layer and facilitates a transformation based on security configuration patterns.

There has been virtually little in the way of user interfaces designed for the exploration and information gathering from large weblog datasets to allow for an integrated and aggregated knowledge collection and information analysis tool. Users have to rely on their own capability to find, select or filter entries and navigate through a blog archive. For weblogs with a large collection of entries this task easily becomes tedious, since current blog interfaces lack fundamental support for facilitating the exploration of their archives. A solution to this problem could be POSTCONNECT, a mature blog-archive visualization tool presented in this paper.

Service-oriented Architectures deliver a flexible infrastructure to allow independently developed software components to communicate in a seamless manner. In the scope of organisational workflows, SOA provides a suitable foundation to execute business processes as an orchestration of multiple independent services. Along with the increased connectivity, the corresponding security risks rise exponentially. However, security requirements are usually defined on a technical level, rather than on an organisational level that would provide a comprehensive view on the participants, the assets and their relationships regarding security. In this paper, we propose an approach to describe security requirements at the business process layer and their translation to concrete security configuration for service-based systems. We introduce security elements for business process modelling which allow to evaluate the trustworthiness of participants based on a rating of enterprise assets and to express security intentions such as confidentiality or integrity on an abstract level. Our aim is to facilitate the generation of security configurations based on the modelled requirements. For this purpose, we foster a model-driven approach: Information at the modelling layer is gathered and translated to a domain-independent security model. Concrete protocols and security mechanisms are resolved based on a security pattern system that is introduced in the course of this paper.

Efficient intrusion detection system (IDS) management is a prominent capability for distributed IDS solutions, which makes it possible to integrate and handle different types of sensors or collect and synthesize alerts generated from multiple hosts located in a loosely coupled environment. Extensibility is the main requirement for most of IDS management systems. The concept of virtualization has been introduced into many popular IDS implementations due to the advantage on isolation and fast recovery in case of being compromised. Advanced capability for combining these newly emerged virtual machine (VM) based IDS approaches is another requirement for IDS management. This paper proposes an extensible IDS management architecture based on a new design of event gatherer component. By using the known IDS standard IDMEF and a plug-in concept, the Event gatherer ensures flexibility and compatibility.Experiments are carried out to demonstrate the extensibility and virtualization-compatibility of the proposed IDS management architecture.

Federated identity management describes a model to enable users to use their digital identities in collaborating companies regardless of organizational borders. The essential pre-requisite to share the user authentication across different security domains is the establishment of trust between the collaborating partners. Usually, this is done by setting up complex contracts, that describe common policies, obligations and procedures to be followed by each collaboration member. The result is a federation, or Circle of Trust, in which each member is willing to trust on assertions made by someone else. Naturally, federations are no isolated structures and members of one federation might also be part of another one - a constellation possible with current federation technologies. However, whether and how the trust relationships of federations can be used to allow access even across multiple federations is a question which has not been answered yet. In this paper, we investigate trust requirements for identity federation topologies. Starting from the classical structure of a Circle of Trust, we go beyond this and identify more complex patterns such as overlapping federations. For each pattern, we identify risks for identity and service providers as well as the necessary trust requirements that must be met to allow such constellations.

The simulation framework VSimRTI provides the integration of multiple heterogeneous simulators to enable V2X application simulations. In order to increase the performance and scalability of complex V2X simulations, a new time management service has been implemented that enables optimistic synchronization in federated simulations. This paper presents the underlying mechanisms as well as the resulting system architecture. To evaluate the improvements that can be achieved, different experiments with highly configurable virtual federates as well as real simulators have been realized. Our experiments show the conditions for an improvement by optimistic synchronization mechanisms and demonstrate the performance increase that can be achieved.

Intrusion detection systems (IDS) have been used widely to detect malicious behaviors in network communication and hosts. IDS management is an important capability for distributed IDS solutions, which makes it possible to integrate and handle different types of sensors or collect and synthesize alerts generated from multiple hosts located in the distributed environment. Facing new application scenarios in cloud computing, the IDS approaches yield several problems since the operator of the IDS should be the user, not the administrator of the cloud infrastructure. Extensibility, efficient management, and compatibility to virtualization-based context need to be introduced into many existing IDS implementations.Additionally, the cloud providers need to enable possibilities to deploy and configure IDS for the user. Within this paper, we summarize several requirements for deploying IDS in the cloud and propose an extensible IDS architecture for being easily used in a distributed cloud infrastructure.

The loosely coupled nature of service-oriented architectures raises the question how information for access control can be managed in an efficient way. Several specifications for Web services exist to describe security requirements and to facilitate a provision of identity information. However, the integration of different standards regarding the expression of identity information in policies, claims and assertions comes along with an increased complexity. In order to identify and address the problems occurring with the combined use of standards as XACML, SAML and WS-Trust, we designed and implemented an architecture for identity- and attribute-based access control in decentralized environments. Our implementation provides an automated generation of access control policies in a format called XACML, a way to communicate required user attributes as claims across different domains based on the standards WS-Trust and WS-Policy, and a consistent mapping of retrieved attribute assertions to the XACML attributes in the access control policy.

Performance is a critical aspect for all kinds of security solutions. The Lock-Keeper is a high-level security solution which implements the concept of physical separation. To evaluate and predict the performance of Lock-Keeper, the internal data transfer process has to be analyzed and simulated. As the Lock-Keeper consists of three independent systems, connected through a hardware-based switch, with file-based communication, the internal data transfer process is relatively complex. In this paper, we propose a mathematical model which can be used to formally represent most of the Lock-Keeper data transfer scenarios. The model is verified by practical experiments for three different application scenarios. We conclude that the performance of the Lock-Keeper system highly depends on the application scenarios.

With a suitable algorithm for ranking the expertise of a user in a collaborative tagging system, we will be able to identify experts and discover useful and relevant resources through them. We propose that the level of expertise of a user with respect to a particular topic is mainly determined by two factors. Firstly, an expert should possess a high quality collection of resources, while the quality of a Web resource depends on the expertise of the users who have assigned tags to it. Secondly, an expert should be one who tends to identify interesting or useful resources before other users do. We propose a graph-based algorithm, SPEAR (SPamming-resistant Expertise Analysis and Ranking), which implements these ideas for ranking users in a folksonomy. We evaluate our method with experiments on data sets collected from Delicious.com comprising over 71,000 Web documents, 0.5 million users and 2 million shared bookmarks. We also show that the algorithm is more resistant to spammers than other methods such as the original HITS algorithm and simple statistical measures.

Vehicle-2-X (V2X) Communication provides the foundation for new applications that enhance both safety and traffic efficiency. Before V2X applications can be deployed in practice, their in-depth analysis is necessary. For this end, detailed and realistic simulations are essential. Depending on the simulated V2X Communication application, particular simulators have to be coupled. For this purpose, we have developed the V2X Simulation Runtime Infrastructure (VSimRTI) offering the flexibility to combine arbitrary simulators. The VSimRTI is derived from concepts of the High Level Architecture (HLA). It synchronizes the simulators and enables the communication among them. Another feature of our simulation environment is the emulation of the environment of V2X Communication applications in real vehicles. As a result, we can integrate real V2X Communication applications without modifications.

The World Wide Web has changed during the last decade. The so-called Web 2.0 enables inexperienced users to become worldwide publishers. Most often these users also don't have any idea about how to protect their own user-generated content or how to trust in content provided by aggregated and syndicated services. Public key infrastructures, digital signatures, and reputation services are well established but hard to understand and to handle for the layperson. We propose an efficient and user-friendly security architecture based on the popular tagging paradigm that connects user-defined tags with security policies, rules, and social network information to ensure access control, data integrity, and confidence also in derived and syndicated data.

In the past decade, the number of digital academic lecture videos has increased dramatically as recording technology has become more affordable. There are technical problems in the use of recorded lectures for learning: the problem of easy access to the multimedia lecture video content and the problem of finding the appropriate information. The first step to a solution is to segment the videos into smaller cohesive areas. In this paper, we present a study on segmenting recorded lecture videos based on their transcripts with standard linear text segmentation algorithm (LTSA). Our evaluation dataset is based on different languages and various speakers' recordings. Three different tests analyze the outcome of ten algorithms: 1) Whether LTSA is able to segment the transcript into the slide transitions. 2) The presentation slides are used as an additional resource for the segmenting procedure. 3) Analyzing the topic boundaries independently from the slide transitions.

The number of digital lecture video recordings has increased dramatically. The accessibility, usability and the traceability of their content for students-use is limited. Therefore retrieval of audiovisual lecture recordings is a complex task. Speech recognition is applied to create a tentative and deficient transcription of the video recordings. The imperfect transcription is sufficient to generate semantic metadata serialized in an OWL file. A question answering system based on the automatically generated semantic annotations and a semantic search engine are presented. The annotation process is discussed, evaluated and compared to a perfectly annotated OWL file and, further, to a corrected transcript of the lecture.

No abstract available.

Access Control decisions are based on the authorisation policies defined for a system as well as observed context and behaviour when evaluating these constraints at runtime. Workflow management systems have been recognised as a primary source for defining authorisation policies at workflow designtime, as well as generating context at runtime.

In this paper, we study and compare three different but related types of metadata about Web documents: social annotations provided by readers of Web documents, hyperlink anchor text provided by authors of Web documents, and search queries of users trying to find Web documents. We introduce a large research data set called CABS120k, which we have created for this study from a variety of information sources such as AOL500k, the Open Directory Project, del.icio.us/Yahoo!, Google and the WWW in general. We use this data set to investigate several characteristics of said metadata including length, novelty, diversity, and similarity and discuss theoretical and practical implications.

Constraints are an important aspect of role-based access control (RBAC) and its different extensions. They are often regarded as one of the principal motivation behind these access control models. In this paper, we introduce two novel authorization constraint specification schemes named as prohibition constraint scheme and obligation constraint scheme. Both of them can be used for expressing and enforcing authorization constraints. These schemes strongly bind to authorization entity set functions and authorization entity relation functions, so they can provide the system designers a clear view about which functions should be defined in an authorization constraint system. Based on these functions, different kinds of constraint schemes can be easily defined. The security administrators can use these functions to create constraint schemes for their day-to-day operations. The constraint system can be scalable through defining new functions. This approach goes beyond the well known separation of duty constraints, and considers many aspects of entity relation constraints.

Web-streaming lectures overcome the space and time barriers between learning and teaching, but bring higher requirements on the learning feedback of students when they browse lectures. In this paper, we discover the students learning interest from their usage data in Web-based learning environment by using multi data mining methods. The learning interests are expressed in six questions, which were asked by the teachers. We use simple statistics, associate rules mining, multi linear regression and similarity comparing to answer different questions. The usage data of online learners are heterogeneous, including HTTP server logs and REAL Helix Universal logs, and these heterogeneous usage data are transformed into students browsing profiles. We implement our work on our Web-based learning environment: tele-TASK. The mined results help teachers to know their students clearly and adjust their teaching schedules efficiently

Collaborative tagging describes the process by which many users add metadata in the form of unstructured keywords to shared content. The recent practical success of web services with such a tagging component like Flickr or del.icio.us has provided a plethora of user-supplied metadata about web content for everyone to leverage.

This paper describes our e-librarian service that understands students' complete questions in natural language and retrieves very few but pertinent learning objects, i.e., short multimedia documents. The system is based on three key components: the formal representation of a domain ontology, a mechanism to automatically identify learning objects out of a knowledge source, and a semantic search engine that yields only pertinent results based on the freely formulated questions in natural language.We report on experiments about students' acceptance to enter complete questions instead of only keywords, and about the benefits of such a virtual personal teacher in an educational environment.

Easy to use, interactive web applications accumulating data from heterogeneous sources represent a recent trend on the World Wide Web, referred to as the Social Web. There however, security standards are often disregarded in favor of interface design or brand new features. This prevents the new services from gaining ground in the enterprise, in medical or e-government environments. We propose the deployment of XML Digital Signatures on web content and demonstrate how an architecture enabling for various security properties would look like. The solution proposed will benefit from the research on security engineering in Service-Oriented Architectures and thus allows for an in-depth analysis on the results.

In this paper, we discover the changes of students' learning interest from their usage data in web-based learning environment. Due to the effects on each other of the changes in Web students and Web lectures, we seek a method that integrates the changes in both sides to measure the changes of learning interest. We implement our work on our Web-based learning environment: tele-TASK. The mined results help teachers to know their students clearly and adjust their teaching schedules efficiently.

To effectively participate in modern collaborations, member organizations must be able to share specific data and functionality with collaboration partners, while ensuring their resources are safe from inappropriate access. This requires access control models, policies, and enforcement mechanisms for the shared resources. This paper specifically addresses how to reduce the information leaks caused by authorization policies used in collaborative computing environment. The basic principle is defining some labels that specify the information flow constraints, and assigning them to authorization policy components. The usages of labeled policy components must obey the information flows constraints defined by the labels in order to avoid authorization policy components being misused. This label can also improve the authorization policy administration.

Summary form only given. OpenSST (Open Simple Secure Transaction) is a free software project and aims to create an efficient, open and secure alternative to the secure proprietary transaction protocol. OpenSST uses XML syntax to specify its message format, and relies on prevalent cryptographic algorithms to secure the transaction. OpenSST is designed to provide user with the guarantee on confidentiality, integrity, authenticity and non-repudiation. One feature of OpenSST's is that it defines a simple message structure so that it can easily be integrated into different usage frameworks. This paper mainly describes OpenSST's logic architecture and message format through an http-based prototype.

Tele-Lab "IT-security" Server is a new concept for practical security education. The Tele-lab server builds a virtual security laboratory using lightweight virtual machines and features a pure Web user interface. Thus, students can easily exercise security over the Internet. Running such a security laboratory on the Internet is difficult. Its infrastructure is subject to misuse and crash due to the nature of security tasks. The management of virtual machines is a crucial component which enables running the Tele-Lab server on the Internet. It effectively saves resources and improves security and reliability of the virtual laboratory. This paper briefly reviews the architecture of the Tele-Lab "IT-security" server, describes its virtual machine management in detail, and presents some experiential and experimental results.

In this paper we describe a mechanism for managing authorisation policies in distributed environments. This mechanism is based on public key infrastructure (PKI) and privilege management infrastructure (PMI). In our approach each domain comprises a root policy and some subordinate authorisation policies. The root policy specifies how to use the subordinate policies. The subordinate policies describe the access control rules that are used for making access control decisions. The subordinate policies can be defined and managed independently and dynamically loaded into the access control system at runtime. All these policies are stored in X.509 attribute certificates (ACs), thus guaranteeing their integrity. The AC that holds root policy is co-located with access control system; the ACs that holds subordinate policies can be distributed. In the root policy we use policy schemes, policy sub-schemes and policy hierarchies to manage the subordinate policies; because they make the policy management flexible and easy.

Providing hands-on experience by live exercises is essential for current IT security education. Therefore, Tele-Lab IT security, a Web-based training system, is being developed at the University of Trier, Germany. It attempts to integrate a security laboratory on the Internet using well-managed virtual machines which allow students to gain experiences of security technologies and tools in a reliable and secure way. We describe user interface and architecture as well as some security considerations.

Dynamic reconfiguration is absorbing more and more research focus for its increasing demand in inconstant distributed application. In This work we propose a routing based workflow to model the dataflow, runtime state and control management of cooperating components. Routing based workflow successfully realizes dynamic reconfiguration by the way of modifying routing structure and simplifies the hard problem of maintaining consistence into rather easy issue of synchronization. A detailed analysis is also given to show the great flexibility for construction of software architecture and potential applications.

IT security education is an important activity in computer science education. The broad range of existing security threats makes it necessary to teach students the principles of IT security as well as to let them gain hands-on experience. In order to enable students to practice IT security anytime anywhere, a novel tutoring system is being developed at the University of Trier, Germany, which allows them to get familiar with security technologies and tools via the Internet. Based on virtual machine technology, users are able to perform exercises on a Linux system instead of in a restricted simulation environment. This paper describes the user interface of the Tele-Lab IT Security, its system architecture and its functional components.

Image computation is the core operation for optimization and formal verification of sequential systems like controllers or protocols. State exploration techniques based on OBDDs use a partitioned representation of the transition relation to keep the OBDD-sizes manageable. This paper presents a new approach that significantly increases the quality of the partitioning of the transition relation of finite state machines. The heuristic has been successfully applied to reachability analysis and symbolic model checking of real life designs, resulting in a significant reduction in CPU time as well as in memory consumption.

Image computation is the core operation for optimization and formal verification of sequential systems like controllers or protocols. State exploration techniques based on ordered binary decision diagrams (OBDDs) use a partitioned representation of the transition relation to keep the OBDD-sizes manageable. This paper presents algorithms for building a hierarchically partitioned transition relation and conjunction scheduling based on this partitioning. The conjunction scheduling algorithm allows one to dynamically reorder partitions and is targeted to improve the AndExist operation. Model checking experiments prove the effectiveness of the new algorithms.

Modern high-quality medicine would be inconceivable without computer and communication technology. A physician can manage an enormous stream of data (especially images) only by using computerized information systems. That's why telemedicine applications are widely accepted in radiology. High standards for patient health care cannot be maintained without the introduction of modern radiological information systems (RISs) and PACS. This paper introduces a new intranet/Internet-oriented RIS for the transmission, visualization and processing of medical images, which can be used in hospitals and in doctor's offices.

Describes the development of a multimedia editor for radiology which allows one to load, store and work with radiological reports. These reports conform to the DICOM standard by using the 'structured reporting' supplement of DICOM. The user interface of the editor is implemented in Java. A DICOM toolkit is used to implement the structure of the reports. The toolkit is implemented in the programming languages C and C++ and can be compiled on the operating systems Linux, Windows NT/9x and Solaris. The programming languages used for the implementation of the user interface and for the construction of DICOM-conforming reports are connected via the Java Native Interface (JNI). The possibility to record spoken language or noises, to visualize DICOM images and to edit natural text is also provided.

This panel focuses on how new Web-based frameworks can enable academic research and development in VLSI CAD. Open-source initiatives, portals for research communities or for the EDA community at large, algorithm benchmarking support, etc. are a few of the possibilities. Each of the three panelists will describe the current status of efforts toward building new Web-based infrastructure for CAD research. The approaches and “mission statements” illustrate the literally unbounded potential for such frameworks to transform how CAD R&D is performed.

According to our definition, the enterprise-level groupware system (EGS) is the Web-based groupware that focuses specifically on addressing some crucial cooperation requirements put forward by Business-to-Business and Business-to Consumer electronic commerce. In this paper, we propose a new approach for constructing EGSs, in which a latest IETF specification: WebDAV is adopted in order to fully unlease the Web's potential in supporting cooperation activities, and at the same time, an enterprise-level platform: J2EE is adopted in order to ensure some enterprise-level features of the system, e.g., scalability, availability, extensibility and security. We then introduce a prototype EGS implementation called "Cooperative Workbench", which is developed in our institute. This prototype has partly proven the advantages of this new approach and its promising application prospects in the future EGSs.

Symbolic Model checking is a widely used technique in sequential verification. As the size of the OBDDs and also the computation time depends on the order of the input variables, the verification may only succeed if a well suited variable order is chosen. Since the characteristics of the represented functions are changing, the variable order has to be adapted dynamically. Unfortunately, dynamic reordering strategies are often very time consuming and sometimes do not provide any improvement of the OBDD representation. This paper presents adaptions of reordering techniques originally intended for combinatorial verification to the specific requirements of symbolic model checking. The techniques are orthogonal in the way that they use either structural information about the OBDDs or semantical information about the represented functions. The application of these techniques substantially accelerates the reordering process and makes it possible to finish computations, that are too time consuming, otherwise.

We discuss a service that aims to make quoting of online documents, "Web contents" easy and provable. For that reason we report the conception of a gateway that works as a trusted third party (TTP) service which is based on a public key infrastructure (PKI). The developed service consists of the signing of any data-transmission that was done via the TTP-gateway. After the data-transfer a set of data can be requested from the used gateway that is signed with the TTP-gateways private key. This signed set of data contains for each request that was processed by the gateway at least three components. Those are the request from the client, the reply from the server and finally the signature of the (TTP) server. Storing this signed data the recipient at the client side can provide it to other parties suitable for a latter verification of the data transfer. The TTP server generates automatically verifiable statements of the kind "this request resulted in that response". Now anyone that trusts the chosen TTP-gateways statements will be able to verify the data-transfer by the use of the trusted third parties certified public key. Furthermore we describe a prototype implementation of such a service using HTTP. Finally a possible employment of the TTP-gateway is discussed.

We propose a new algorithm, called linear sifting, for the optimization of decision diagrams that combines the efficiency of sifting and the power of linear transformations. The new algorithm is applicable to large examples, and in many cases leads to substantially more compact diagrams when compared to simple variable reordering. We also show in what sense linear transformations complement variable reordering and how the technique can be applied to verification issues. Going a step further, we discuss a synthesis scenario where-due to the complexity of the target function-it is inevitable to decompose the function in a preprocessing step. By using linear sifting it is possible to extract a linear filter and, hence, to achieve the necessary decomposition. Using this method we were able to synthesize functions with standard tools which fail otherwise.

Variable reordering is the main approach to minimize the size of Ordered Binary Decision Diagrams. But despite the huge effort spent, up to now, to design different reordering heuristics, their performance often does not meet the needs of the applications. In many OBDD-based computations, the time cost for reordering dominates the time spent by the computation itself. There are some known approaches for accelerating the reordering by taking advantage of structural properties of OBDDs and functions represented. In this paper, we propose a reordering method that exploits application specific information. The main idea is to drive the reordering process by the computation. This effects an acceleration of the whole computation rather than of the reordering only. The power of the approach is illustrated by speeding up forward traversal of finite state machines.

In this paper a network algorithm to compress and to reconstruct DICOM-images is presented. The algorithm consists of three parts: the bit-levels of the original image is classified on the basis of a temporary segmentation and analysis of the image structure and afterwards divided into two parts. The first part of the image consists of the highest bit-levels of the original image, the second one contains the rest of the original image. Depending on the segment analyst of the 2nd image the segment orientation is defined and the 2nd image is split in two areas. The first area is the object area, the second is the unusable area. Finally the first part of the image is coded by the LZW-method. The direction of code depends on the orientation of the segments. The region of the object of the 2nd image is JPEG coded. The results of a statistic model of this algorithm is also shown. The algorithm is implemented in the JAVA language.

In this paper a complete concept will be shown, to build a cheap, modular internet/intranet-based radiologic information system for hospitals or settled MDs. The goal of this system is to achieve internal/external access to DICOM-data, which are produced and archived in the several image-producing departments. Using the DICOM-standard as well as modern, platform independent Java programming and internet/intranet technology this goal can be reached. Core of the whole system is a PACS. Integration of all digital modalities by the DICOM standard is an essential proposal of the system's concept.

Model checking has been proven to be a powerful tool in verification of sequential circuits, reactive systems, protocols, etc. The model checking of systems with huge state spaces is possible only if there is a very efficient representation of the model. Ordered Binary Decision Diagrams (OBDDs) allow an efficient symbolic representation of the model. Our goal is to accelerate the variable reordering process but retaining good OBDD sizes. To obtain this, we adapted two methods introduced by Meinel and Slobodova called Block Restricted Sifting (BRS) and Sample Sifting to the needs of model checking.

In order to simplify a synthesis task for particularly hard functions it is sometimes inevitable to decompose the function in a preprocessing step. We propose a new algorithm for automatically decomposing a target function by extracting a linear filter within the synthesis process. The algorithm is an application of the Linear Sifting algorithm which has been proposed in Meinel et al. (1996). Using this method we were able to synthesize functions with standard tools which fail otherwise.

We propose a new algorithm, called linear sifting, for theoptimization of decision diagrams that combines the efficiency of sifting and the power of linear transformations. We show that the new algorithm is applicable to large examples, and that inmany cases it leads to substantiallymore compact diagrams when compared to simple variablereordering. We show inwhat sense linear transformationscomplement variable reordering, and we discuss applications of the new technique to synthesis and verification.

The use of Ordered Binary Decision Diagrams (OB-DDs) as a representation of Boolean functions brought essential progress in many different applications. The optimization of the OBDD-size by the choice of the variable ordering is known to be NP-hard. The known heuristics for finding an initial ordering and for reordering still have insufficient performance. Rudell's sifting is one of the most successful reordering algorithms that is application independent and can be used dynamically. In this paper, we propose a method based on some communication complexity considerations that improves the time performance of the sifting. The main idea is to restrict the reordering of variables to blocks that are determined according to readily computable OBDD-measure. Experimental comparison of the proposed block-restricted sifting with the original algorithm shows a speed-up by factor two without any loss in the final size.

Reduced ordered binary decision diagrams (OBDD's) are nowadays the state-of-the-art representation scheme for Boolean functions in Boolean manipulation. Recent results have shown that it is possible to use the more general concept of free binary decision diagrams (FBDD's) without giving up most of the useful computational properties of OBDD's, but possibly reducing the space requirements considerably. The amount of space reduction depends essentially on the shape of so-called FBDD-types the Boolean manipulation in terms of FBDD's is based on. Here, we propose some heuristics for deriving tree-like FBDD-types from given circuit descriptions. The experimental results we obtained clearly demonstrate that the FBDD-approach is not only of theoretical interest, but also of practical usefulness even in the ease of using merely such simple-structured tree-based FBDD-types as produced by the investigated heuristics.

It is well-known that the size of an ordered binary decision diagram (OBDD) may depend crucially on the order in which the variables occur. In the paper, we describe an implementation of an output-efficient algorithm that transforms an OBDD P representing a Boolean function f with respect to one variable ordering /spl pi/ into an OBDD Q that represents f with respect to another variable ordering /spl sigma/. The algorithm runs in average time O(|P/spl par/Q|) and requires O(|P|+|Q|) space. The importance of the algorithm is demonstrated by means of experimental results on basically two different applications. In one of them, the algorithm is used merely once. Such transformations are needed to test equivalence or to perform synthesis on OBDD's in which variables appear in different orders. The other application shows a way how to decrease the size of intermediate OBDD representations of a given circuit in the course of its symbolic simulation. Here, the algorithm is used dynamically, whenever the size of the manipulated OBDD's becomes too large.

We present the concept of TBDD's which considerably enlarges the class of Boolean functions that can be efficiently manipulated in terms of OBDD's. It extends the idea of using domain trans-formations, which is well-known in many areas of mathematics, physics, and technical sciences, to the context of OBDD-based Boolean function manipulation in CAD: Instead of working with the OBDD-representation of a function f, TBDD's allow working with an OBDD-representation of a suited cube transformed version of f. Besides of giving some theoretical insights into the new concept, we investigate in some detail cube transformations which are based on complete types. We show that such TBDD-representations can be derived similarly as OBDD-representations, give evidence of the practical importance of such TBDD's by presenting very small-size TBDD-representations of the hidden weighted bit functions

OBDD's are the state-of-the-art data structure for Boolean function manipulation. Basic tasks of Boolean manipulation such as equivalence test, satisfiability test, tautology test and single Boolean synthesis steps can be performed efficiently in terms of fixed ordered OBDD's. The bottleneck of most OBDD-applications is the size of the represented Boolean functions since the total computation merely remains tractable as long as the OBDD-representations remain of reasonable size. Since it is well known that OBDD's are restricted FBDD's (free BDD's, i.e., BDD's that test, on each path, each input variable at most once), and that FBDD-representations are often much more (sometimes even exponentially more) concise than OBDD-representations. We propose to work with a more general FBDD-based data structure. We show that FBDD's of a fixed type provide, similar as OBDD's of a fixed variable ordering, canonical representations of Boolean functions, and that basic tasks of Boolean manipulation can be performed in terms of fixed typed FBDD's similarly efficient as in terms of fixed ordered OBDD's. In order to demonstrate the power of the FBDD-concept we show that the verification of the circuit design for the hidden weighted bit function proposed Bryant can be carried out efficiently in terms of FBDD's while this is, for principal reasons, impossible in terms of OBDD's.>