In real-world data integration scenarios, traditional equi-joins and other join techniques have huge difficulties due to heterogenity and inconsistencies in attribute values. To address this challenge, we present AutoStarJoin, a technique for automated joins specifically designed for star-join scenarios. The core contribution of our approach is the automated detection of join attributes across arbitrary schemas, reducing or even eliminating the need for manual specification. Our approach analyses first the edit-based distance measures, transforming similar string values within join attributes to facilitate matching, and explores then token-based distance measures, refining the join process by identifying optimal attribute pairs. We evaluate the effectiveness of various distance metrics in terms of join quality and computational efficiency across diverse datasets. Our goal is to generalize across different join scenarios without requiring domain-specific parameter tuning. This level of automation makes our approach suitable for integration into AutoML pipelines where minimal human intervention is desired. However, the correct choice of parameters per data set is crucial. We intend to implement hyperparameter optimization in further research.

An evaluation of the responsibility of generative AI models presents unique challenges that require holistic and practical solutions. This paper introduces an enhanced version of the VERIFAI framework, which extends beyond classification models to assess generative language models as well in terms of ethics, explainability, privacy, and security. Unlike existing theoretical frameworks, VERIFAI provides an integrated, software-driven approach that automates evaluations, ensures reproducibility, and offers actionable insights. To demonstrate its capabilities, we conduct an evaluation of the generative language model Llama-3.2-1B using the Regard metric, which quantifies bias in text generation. Our findings highlight systematic biases in model outputs, reinforcing the need for structured Responsible AI assessments. This work underscores VERIFAI’s scalability, intuitive UI, and advanced analysis capabilities, positioning it as a practical tool for the responsible evaluation of AI models.

The growing complexity of deep learning models introduces challenges in scalability and adaptability. This paper explores how modular design, inspired by software engineering, can enhance deep learning systems. This paper also showed that modern deep learning techniques such as Mixture of Experts (MoE) and LoRA are advancing toward higher modularity. By promoting modular architectures, we emphasize the need to move beyond monolithic models toward more reusable, maintainable, and scalable AI systems, providing a potential research direction for future work.

The early and automatic detection of faulty behavior is essential for maintaining the reliability of a cyber-physical system. In this paper we describe a fault localization approach for such a highly complex distributed system, the optical synchronization system of the European X-ray free-electron laser. Using a dependency graph, we model the relationships between the components and the influences of environmental effects. After we first resolve linear long-term dependencies between dependent components with a correlation analysis, we then use an unsupervised fault detection pipeline consisting of statistical feature extraction and unsupervised anomaly detection to accurately identify anomalies and localize their origins in the system.

This paper introduces Global Contextualized Representations (GCoRe) – an extension for existing transformer-based language models. GCoRe addresses limitations in capturing global context and long-range dependencies by utilizing Graph Neural Networks for graph inference on a context graph constructed from the input text. Global contextualized features, derived from the context graph, are added to the token representations from the base language model. Experiment results show that GCoRe improves the performance of the baseline model (DeBERTa v3) by 0.57% on the HotpotQA dataset and by 0.15% on the SQuAD v2 dataset. In addition, GCoRe is able to answer questions that require logical reasoning and multi-hop inference, while the baseline model fails to provide correct answers.

Our investigation seeks to enhance the understanding of responsible artificial intelligence. The EU is deeply engaged in discussions concerning AI trustworthiness and has released several relevant documents. It’s crucial to remember that while AI offers immense benefits, it also poses risks, necessitating global oversight. Moreover, there’s a need for a framework that helps enterprises align their AI development with these international standards. This research will aid both policymakers and AI developers in anticipating future challenges and prioritizing their efforts. In our study, we delve into the essence of responsible AI and, to our understanding, introduce a comprehensive definition of the term. Through a thorough literature review, we pinpoint the prevailing trends surrounding responsible AI. Using insights from our analysis, we’ve also deliberated on a prospective framework for responsible AI. Our findings emphasize that human-centeredness should prioritized. This entails adopting AI techniques that prioritize ethical considerations, explainability of models, and aspects like privacy, security, and trustworthiness.

The successful operation of the laser-based synchronization system of the European X-Ray Free Electron Laser relies on the precise functionality of numerous dynamic systems operating within closed loops with controllers. In this paper, we present how data-based machine learning methods can detect and classify disturbances to such dynamic systems based on the controller output signal. We present 4 feature extraction methods based on statistics in the time domain, statistics in the frequency domain, characteristics of spectral peaks, and the autoencoder latent space representation of the frequency domain. These feature extraction methods require no system knowledge and can easily be transferred to other dynamic systems. We combine feature extraction, fault detection, and fault classification into a comprehensive and fully automated condition monitoring pipeline. For that, we systematically compare the performance of 19 state-of-the-art fault detection and 4 classification algorithms to decide which combination of feature extraction and fault detection or classification algorithm is most appropriate to model the condition of an actively controlled phase-locked laser oscillator. Our experimental evaluation shows the effectiveness of clustering algorithms, showcasing their strong suitability in detecting perturbed system conditions. Furthermore, in our evaluation, the support vector machine proves to be the most suitable for classifying the various disturbances.

This paper presents a reinforcement learning approach for foreign exchange trading. Inspired by technical analysis methods, this approach makes use of technical indicators by encoding them into Gramian Angular Fields and searches for patterns that indicate price movements using convolutional neural networks (CNN). In addition to the policy that determines the action to take, an extra regression head is utilized to determine the size of market orders. This paper also experimentally shows that maximizing the return of individual trade or cumulative reward in a finite time window results to better performance.

The ongoing COVID-19 pandemic brings new challenges and risks in various areas of our lives. The lack of viable treatments is one of the issues in coping with the pandemic. Developing a new drug usually takes 10-15 years, which is an issue since treatments for COVID-19 are required now. As an alternative to developing new drugs, the repurposing of existing drugs has been proposed. One of the scientific methods that can be used for drug repurposing is literature-based discovery (LBD). LBD uncovers hidden knowledge in the scientific literature and has already successfully been used for drug repurposing in the past. We provide an overview of existing LBD methods that can be utilized to search for new COVID-19 treatments. Furthermore, we compare the three LBD systems Arrowsmith, BITOLA, and SemBT, concerning their suitability for this task. Our research shows that semantic models appear to be the most suitable for drug repurposing. Nevertheless, Arrowsmith currently yields the best results, despite using a co-occurrence model instead of a semantic model. However, it achieves the good results because BITOLA and SemBT currently do not allow for COVID-19 related searches. Once this limitation is removed, SemBT, which uses a semantic model, will be the better choice for the task.

This work introduces Multi-Fusion Network for human-object interaction detection with multiple cameras. We present a concept and implementation of the architecture for a beverage refrigerator with multiple cameras as proof-of-concept. We also introduce an effective approach for minimizing the required amount of training data for the network as well as reducing the risk of overfitting, especially when dealing with a small data set that is commonly recorded by a person or small organization. The model achieved high test accuracy and comparable results in a real-world scenario at the Event Solutions in Hamburg 2019. Multi-Fusion Network is easy to scale due to shared learnable parameters. It is also lightweight, hence suitable to run on small devices with average computation capability. Furthermore, it can be used for smart home applications, gaming experiences, or mixed reality applications.

Damage caused by natural hazards is increasing all over the world. All countries intensify their efforts to predict and prevent hazard events and to decrease disaster impact. Disaster management is one of the challenging, complex and critical application areas dealing with hyper dynamic situation changes, high velocity, voluminous data and organizational heterogeneity. Successful management of disaster response requires flexible and adaptable solution techniques including very accurate, fast and dynamic activity guidance for supporting of process coordination, decision making and information logistics in real-time.

Workflow management systems provide support for structured processes and help to follow the defined business process. Although their importance has been proved by various applications over the last decades they are not appropriate for all use cases. Such workflow management systems are only applicable for domains where the process is well structured and static. In various domains it is essential that the workflow is adapted to the current situation. In this case the traditional workflow systems are not applicable. A flexible approach is required.

Fast databases are no longer nice-to-have – they are a necessity. Many modern applications are becoming performance critical. At the same time, the size of some databases has been increasing to levels that cannot be well supported by current technology. Performance engineering is now becoming a buzzword for database systems. At first physical and partially logical tuning methods have been used for support of high performance systems, but they are mainly based on large and not well understood performance and tuning parameters. Nowadays it becomes obvious that we need methods for systematic performance design.