High-definition (HD) maps are crucial to autonomous driving, providing structured representations of road elements to support navigation and planning. However, existing query-based methods often employ random query initialization and depend on implicit temporal modeling, which lead to temporal inconsistencies and instabilities during the construction of a global map. To overcome these challenges, we introduce a novel end-to-end framework for consistent online HD vectorized map construction, which jointly performs map instance tracking and short-term prediction. First, we propose a Semantic-Aware Query Generator that initializes queries with spatially aligned semantic masks to capture scene-level context globally. Next, we design a History Rasterized Map Memory to store fine-grained instance-level maps for each tracked instance, enabling explicit historical priors. A History-Map Guidance Module then integrates rasterized map information into track queries, improving temporal continuity. Finally, we propose a Short-Term Future Guidance module to forecast the immediate motion of map instances based on the stored history trajectories. These predicted future locations serve as hints for tracked instances to further avoid implausible predictions and keep temporal consistency. Extensive experiments on the nuScenes and Argoverse2 datasets demonstrate that our proposed method outperforms state-of-the-art (SOTA) methods with good efficiency.

The remarkable image generation capabilities of state-of-the-art diffusion models, such as Stable Diffusion, can also be misused to spread misinformation and plagiarize copyrighted materials. To mitigate the potential risks associated with image editing, current image protection methods rely on adding imperceptible perturbations to images to obstruct diffusion-based editing. A fully successful protection for an image implies that the output of editing attempts is an undesirable, noisy image which is completely unrelated to the reference image. In our experiments with various perturbation-based image protection methods across multiple domains (natural scene images and artworks) and editing tasks (image-to-image generation and style editing), we discover that such protection does not achieve this goal completely. In most scenarios, diffusion-based editing of protected images generates a desirable output image which adheres precisely to the guidance prompt. Our findings suggest that adding noise to images may paradoxically increase their association with given text prompts during the generation process, leading to unintended consequences such as better resultant edits. Hence, we argue that perturbation-based methods may not provide a sufficient solution for robust image protection against diffusion-based editing.1

Recent video semantic segmentation (VSS) methods have demonstrated promising results in well-lit environments. However, their performance significantly drops in low-light scenarios due to limited visibility and reduced contextual details. In addition, unfavorable low-light conditions make it harder to incorporate temporal consistency across video frames and thus, lead to video flickering effects. Compared with conventional cameras, event cameras can capture motion dynamics, filter out temporal-redundant information, and are robust to lighting conditions. To this end, we propose EVSNet, a lightweight framework that leverages event modality to guide the learning of a unified illumination-invariant representation. Specifically, we leverage a Motion Extraction Module to extract short-term and long-term temporal motions from event modality and a Motion Fusion Module to integrate image features and motion features adaptively. Furthermore, we use a Temporal Decoder to exploit video contexts and generate segmentation predictions. Such designs in EVSNet result in a lightweight architecture while achieving SOTA performance. Experimental results on 3 large-scale datasets demonstrate our proposed EVSNet outperforms SOTA methods with up to 11 × higher parameter efficiency.

3D human pose estimation (3D HPE) is an important computer vision task with various practical applications. Researchers have proposed various deep learning-based methods for 3D HPE. However, the majority of such methods rely on lifting 2D pose sequence to 3D which do not perform well in challenging scenarios and are often computationally expensive. Such methods typically rely on 2D joint coordinates which do not provide much spatial context to solve ambiguity problem. In addition, merely relying on information extracted from RGB frames may miss temporal information and structural context. Thus, in this paper, we propose a framework that incorporates event stream as an additional input since event features provide such useful information. Moreover, instead of using 2D joint coordinates in pose sequence, our framework uses intermediate visual representations produced by off-the-shelf 2D pose detectors to implicitly encode joint-centric spatial context. Our new framework is a novel state space model (SSM)-based solution called Event-Guided Context Aware MambaPose (CA-MambaPose). In CA-MambaPose framework, we design a novel cross modality fusion mamba module to skillfully fuse the RGB and Event features. CA-MambaPose has lower computational cost due to the efficiency of Mamba blocks. We conduct extensive experiments to evaluate CA-MambaPose using two existing datasets. Our experimental results show that CA-MambaPose achieves better performance than SOTA methods.

3D human pose estimation (3D HPE) is an important computer vision task with various practical applications. However, 3D pose estimation for multi-person from a monocular video (3DMPPE) is particularly challenging. Recent transformer-based approaches focus on capturing the spatial-temporal information from sequential 2D poses, which unfortunately loses the visual feature relevant for 3D pose estimation. In this paper, we propose an end-to-end framework called Event Guided Video Transformer (EVT) which predicts 3D poses directly from video frames by learning spatial-temporal contextual information from visual features effectively. In addition, our design is the first that incorporates event features to help guide 3D pose estimation. EVT first decouples persons into different instance-aware feature maps from video frames. These features containing specific clues of body structure information are then fed together with event features into an attention based Event-Aware Embedding Module. Next, the fused features for each instance are then fed into an intra-human relation extraction module and subsequently to a temporal transformer to extract inter-frame relationship. Finally, the extracted features are fed into a decoder for 3D pose estimation. Experiments using three widely used 3D pose estimation benchmarks show that our proposed EVT achieves better performance than state-of-the-art models.

Routine visual inspections of concrete structures are imperative for upholding the safety and integrity of critical infrastructure. Such visual inspections sometimes happen under low-light conditions, e.g., checking for bridge health. Crack segmentation under such conditions is challenging due to the poor contrast between cracks and their surroundings. However, most deep learning methods are designed for well-illuminated crack images and hence their performance drops dramatically in low-light scenes. In addition, conventional approaches require many annotated low-light crack images which is time-consuming. In this paper, we address these challenges by proposing CrackNex, a framework that utilizes reflectance information based on Retinex Theory to learn a unified illumination-invariant representation. Furthermore, we utilize few-shot segmentation to solve the inefficient training data problem. In CrackNex, both a support prototype and a reflectance prototype are extracted from the support set. Then, a prototype fusion module is designed to integrate the features from both prototypes. CrackNex outperforms the SOTA methods on multiple datasets. Additionally, we present the first benchmark dataset, LCSD, for low-light crack segmentation. LCSD consists of 102 well-illuminated crack images and 41 low-light crack images. The dataset and code are available at https://github.com/zy1296/CrackNex.

Many learning-based low light image enhancement (LLIE) algorithms are based on the Retinex theory. However, the Retinex-based decomposition models introduce corruptions which limit their enhancement performance. In this paper, we propose a Latent Disentangle-based Enhancement Network (LDE-Net) for low light vision tasks. The latent disentanglement module disentangles the input image in latent space such that no corruption remains in the disentangled Content and Illumination components. For LLIE task, we design a Content-Aware Embedding (CAE) module that utilizes Content features to direct the enhancement of the Illumination component. For downstream tasks (e.g. nighttime UAV tracking and low light object detection), we develop an effective light-weight enhancer based on the latent disentanglement framework. Comprehensive quantitative and qualitative experiments demonstrate that our LDE-Net significantly outperforms state-of-the-art methods on various LLIE benchmarks. In addition, the great results obtained by applying our framework on the downstream tasks also demonstrate the usefulness of our latent disentanglement design.

For autonomous vehicles (AVs), the ability for effective end-to-end perception and future trajectory prediction is critical in planning a safe automatic maneuver. In the current AVs systems, perception and prediction are two separate modules. The prediction module receives only a restricted amount of information from the perception module. Furthermore, perception errors will propagate into the prediction module, ultimately having a negative impact on the accuracy of the prediction results. In this paper, we present a novel framework termed BEV-TP, a visual context-guided center-based transformer network for joint 3D perception and trajectory prediction. BEV-TP exploits visual information from consecutive multi-view images and context information from HD semantic maps, to predict better objects’ centers whose locations are then used to query visual features and context features via the attention mechanism. Generated agent queries and map queries facilitate learning of the transformer module for further feature aggregation. Finally, multiple regression heads are used to perform 3D bounding box detection and future velocity prediction. This center-based approach achieves a differentiable, simple, and efficient E2E trajectory prediction framework. Extensive experiments conducted on the nuScenes dataset demonstrate the effectiveness of BEV-TP over traditional pipelines with sequential paradigms.

In recent years, the design of models for performing the trajectory prediction task, one of the critical tasks in autonomous driving, has received great attention from researchers. However, accurately predicting future locations is challenging due to the difficulty of learning accurate intentions and modeling multimodality. Historical paths at a certain location can help predict the future trajectory of an agent currently located in that position and address these limitations. In this work, we propose a goal-based local behavior guided model, Goal-LBP, using such information (referred to as local behavior data) to generate potential goals and guide the prediction of trajectories conditioned on such goals. Goal-LBP uses Transformer encoders to extract homogeneous features and attention mechanism to represent the heterogeneous interactions and subsequently uses an encoder-decoder Gated Recurrent Unit (GRU) model to generate predictions. We evaluate our Goal-LBP using two large-scale real-world autonomous driving datasets, namely nuScenes and Argoverse. Our results show that compared to several SOTA models, Goal-LBP achieves the best ADE/FDE performance and it ranked #2 on the leaderboard of the nuScenes trajectory benchmark in June 2023. In addition, we also demonstrate that our local behavior estimator block can be easily added to two existing SOTA methods, namely AgentFormer and LaPred. Adding this LBE block improves the original AgentFormer and LaPred performance by at least 10%.

Trajectory Prediction (TP) is a critical component in the control system of an Autonomous Vehicle (AV). It predicts future motion of traffic agents based on observations of their past trajectories. Existing works have studied the vulnerability of TP models when the perception systems are under attacks and proposed corresponding mitigation schemes. Recent TP designs have incorporated context map information for performance enhancements. Such designs are subjected to a new type of attacks where an attacker can interfere with these TP models by attacking the context maps. In this paper, we study the robustness of TP models under our newly proposed map-based adversarial attacks. We show that such attacks can compromise state-of-the-art TP models that use either image-based or node-based map representation while keeping the adversarial examples imperceptible. We also demonstrate that our attacks can still be launched under the black-box settings without any knowledge of the TP models running underneath. Our experiments on the NuScene dataset show that the proposed map-based attacks can increase the trajectory prediction errors by 29-110%. Finally, we demonstrate that two defense mechanisms are effective in defending against such map-based attacks.

Even though memory denial-of-service attacks can cause severe performance degradations onco-locatedvirtual machines, a previous detection scheme against such attacks cannot accurately detect the attacks and also generates high detection delay and high performance overhead since it assumes that cache-related statistics of an application follow the same probability distribution at all times, which may not be true for all types of applications. In this paper, we present the experimental results showing the impacts of memory DoS attacks on different types of cloud-based applications. Based on these results, we propose two lightweight and responsive Statistical based Detection Schemes (SDS/B and SDS/P) that can detect such attacks accurately. SDS/B constructs a profile of normal range of cache-related statistics for all applications and use statistical methods to infer an attack when the real-time collected statistics exceed this normal range, while SDS/P exploits the increased periods of access patterns for periodic applications to infer an attack. Upon SDS, we further leverage deep neural network (DNN) techniques to design a DNN-based detection scheme that is general to various types of applications and more robust to adaptive attack scenarios. Our evaluation results show that SDS/B, SDS/P and DNN outperform the state-of-the-art detection scheme, e.g., with 65% higher specificity, 40% shorter detection delay, and 7% less performance overhead. We also discuss how to use SDS and DNN-based detection schemes under different situations.

The task of instance segmentation in videos aims to consistently identify objects at pixel level throughout the entire video sequence. Existing state-of-the-art methods either follow the tracking-by-detection paradigm to employ multi-stage pipelines or directly train a complex deep model to process the entire video clips as 3D volumes. However, these methods are typically slow and resource-consuming such that they are often limited to offline processing. In this paper, we propose SRNet, a simple and efficient framework for joint segmentation and tracking of object instances in videos. The key to achieving both high efficiency and accuracy in our framework is to formulate the instance segmentation and tracking problem into a unified spatial-relation learning task where each pixel in the current frame relates to its object center, and each object center relates to its location in the previous frame. This unified learning framework allows our framework to perform join instance segmentation and tracking through a single stage while maintaining low overheads among different learning tasks. Our proposed framework can handle two different task settings and demonstrates comparable performance with state-of-the-art methods on two different benchmarks while running significantly faster.

Training a semantic segmentation model requires large densely-annotated image datasets that are costly to obtain. Once the training is done, it is also difficult to add new object categories to such segmentation models. In this paper, we tackle the few-shot semantic segmentation problem, which aims to perform image segmentation task on unseen object categories merely based on one or a few support example(s). The key to solving this few-shot segmentation problem lies in effectively utilizing object information from support examples to separate target objects from the background in a query image. While existing methods typically generate object-level representations by averaging local features in support images, we demonstrate that such object representations are typically noisy and less distinguishing. To solve this problem, we design an object representation generator (ORG) module which can effectively aggregate local object features from support im- age(s) and produce better object-level representation. The ORG module can be embedded into the network and trained end-to-end in a weakly-supervised fashion without extra human annotation. We incorporate this design into a modified encoder-decoder network to present a powerful and efficient framework for few-shot semantic segmentation. Experimental results on the Pascal-VOC and MS-COCO datasets show that our approach achieves better performance compared to existing methods under both one-shot and five-shot settings.

In this poster, we present measurement studies of the impact of memory DoS attacks on different types of cloud-based applications. Based on the observations, we propose a lightweight, responsive Statistical based Detection Scheme (SDS) that can detect such attacks accurately. Our initial evaluation results show that SDS outperforms the state-of-the-art detection scheme up to 2% higher recall, up to 65% higher specificity, and up to 40% shorter detection delay.

Even though memory-based denial-of-service attacks can cause severe performance degradations on co-located virtual machines, a previous detection scheme against such attacks cannot accurately detect the attacks and also generates high detection delay and high performance overhead since it assumes that cache-related statistics of an application follow the same probability distribution at all times, which may not be true for all types of applications. In this paper, we present the experimental results showing the impacts of memory DoS attacks on different types of cloud-based applications. Based on these results, we propose two lightweight, responsive Statistical based Detection Schemes (SDS/B and SDS/P) that can detect such attacks accurately. SDS/B constructs a profile of normal range of cache-related statistics for all applications and use statistical methods to infer an attack when the real-time collected statistics exceed this normal range, while SDS/P exploits the increased periods of access patterns for periodic applications to infer an attack. Our evaluation results show that SDS/B and SDS/P outperform the state-of-the-art detection scheme, e.g., with 65% higher specificity, 40% shorter detection delay, and 7% less performance overhead.

The explosive usage of mobile devices enables conducting electronic transactions involving direct signature on such devices. Thus, user signature verification becomes critical to ensure the success deployment of online transactions such as approving legal documents and authenticating financial transactions. Existing approaches mainly focus on user verification targeting the unlocking of mobile devices or performing continuous verification based on a user's behavioral traits. Few studies provide efficient real-time user signature verification. In this work, we propose a critical segment based online signature verification system to secure mobile transactions on multi-touch mobile devices. Our system identifies and exploits the segments which remain invariant within a user's signature to capture the intrinsic signing behavior embedded in each user's signature. Our system extracts useful features from a user's signature that describe both the geometric layout of the signature as well as behavioral and physiological characteristics in the user's signing process. Given the input signatures for user enrollment, our system further designs a quality score to identify the problematic signature sets to achieve robust user signature profile construction. Moreover, we develop the signature normalization and interpolation methods to achieve robust signature verification in the presence of signature geometric distortions caused by different writing sizes, orientations and locations on touch screens. Our experimental evaluation of 25 subjects over six months time period shows that our system is highly accurate in provide signature verification and robust to signature forging attacks.

Deploying a deep learning model on mobile/IoT devices is a challenging task. The difficulty lies in the trade-off between computation speed and accuracy. A complex deep learning model with high accuracy runs slowly on resource-limited devices, while a light-weight model that runs much faster loses accuracy. In this paper, we propose a novel decomposition method, namely DAC, that is capable of factorizing an ordinary convolutional layer into two layers with much fewer parameters. DAC computes the corresponding weights for the newly generated layers directly from the weights of the original convolutional layer. Thus, no training (or fine-tuning) or any data is needed. The experimental results show that DAC reduces a large number of floating-point operations (FLOPs) while maintaining high accuracy of a pre-trained model. If 2% accuracy drop is acceptable, DAC saves 53% FLOPs of VGG16 image classification model on ImageNet dataset, 29% FLOPS of SSD300 object detection model on PASCAL VOC2007 dataset, and 46% FLOPS of a multi-person pose estimation model on Microsoft COCO dataset. Compared to other existing decomposition methods, DAC achieves better performance.

In recent years, robotic technologies, e.g. drones or autonomous cars have been applied to the agricultural sectors to improve the efficiency of typical agricultural operations. Some agricultural tasks that are ideal for robotic automation are yield estimation and robotic harvesting. For these applications, an accurate and reliable image-based detection system is critically important. In this work, we present a low-cost strawberry detection system based on convolutional neural networks. Ablation studies are presented to validate the choice of hyper-parameters, framework, and network structure. Additional modifications to both the training data and network structure that improve precision and execution speed, e.g., input compression, image tiling, color masking, and network compression, are discussed. Finally, we present a final network implementation on a Raspberry Pi 3B that demonstrates a detection speed of 1.63 frames per second and an average precision of 0.842.

Obesity, a serious public health concern worldwide, increases the risk of many diseases, including hypertension, stroke, and type 2 diabetes. To tackle this problem, researchers collect diverse types of data, which includes biomedical, behavioral and activity, and utilize machine learning techniques to mine hidden patterns for obesity status improvement prediction. While existing machine learning methods such as Recurrent Neural Networks (RNNs) provide exceptional results, it is challenging to discover hidden patterns of the sequential data due to the irregular observation time instances. Meanwhile, the lack of understanding of why those learning models are effective also limits further improvements on their architectures. Thus, we develop a RNN based time-aware architecture to handle irregular observation times and identify relevant feature extractions from longitudinal patient records for obesity status improvement pre-diction. Evaluations of real-world data involving activity data collected from wearables and electronic health records demonstrate that our proposed method can capture the underlying structures in users' time sequences with irregularities, and achieve an accuracy of 77% in predicting the obesity status improvement.

Modern face recognition systems are accurate but they are vulnerable to different types of spoofing attacks. To solve this problem, conventional face authentication systems typically employ an additional module to analyze the liveness of the input faces before feeding it into the face recognition module. Such two-stage designs not only suffer from longer processing time but also require more storage and resources, which are usually limited on mobile and embedded platforms. In this paper, we propose a multi-task Convolutional Neural Network(CNN), namely LiveFace, for face-authentication. Given an input face image, LiveFace generates two outputs through a single stage: (i) a face representation that can be used for identification or verification, and (ii) the corresponding liveness score. The two tasks share lower layers to reduce the computation cost. Experimental results using three datasets show that our model achieves a comparable performance on both face recognition and anti-spoofing tasks but much faster than conventional authentication systems. In addition, we have implemented a prototype of our scheme on Android phones and demonstrated that our scheme can run in real-time on three Android devices that we have tested.

Designing a scheme that can achieve a good performance in predicting single person activities and group activities is a challenging task. In this paper, we propose a novel robust and efficient human activity recognition scheme called ReHAR, which can be used to handle single person activities and group activities prediction. First, we generate an optical flow image for each video frame. Then, both video frames and their corresponding optical flow images are fed into a Single Frame Representation Model to generate representations. Finally, an LSTM is used to predict the final activities based on the generated representations. The whole model is trained end-to-end to allow meaningful representations to be generated for the final activity recognition. We evaluate ReHAR using two well-known datasets: the NCAA Basketball Dataset and the UCFSports Action Dataset. The experimental results show that the proposed ReHAR achieves a higher activity recognition accuracy with an order of magnitude shorter computation time compared to the state-of-the-art methods.

Activity recognition has become an important function in many emerging computer vision applications e.g. automatic video surveillance system, human-computer interaction application, and video recommendation system, etc. In this paper, we propose a novel semantics based group activity recognition scheme, namely SBGAR, which achieves higher accuracy and efficiency than existing group activity recognition methods. SBGAR consists of two stages: in stage I, we use a LSTM model to generate a caption for each video frame; in stage II, another LSTM model is trained to predict the final activity categories based on these generated captions. We evaluate SBGAR using two well-known datasets: the Collective Activity Dataset and the Volleyball Dataset. Our experimental results show that SBGAR improves the group activity recognition accuracy with shorter computation time compared to the state-of-the-art methods.

Until today, finding the most suitable APIs to use in an application was burdensome, requiring manual and time-consuming searches across a diverse set of websites, in particular regarding how multiple APIs could be combined and worked together (i.e. API mashups). In this paper, we propose a new method to automatically generate API mashups through real-world data collection, text mining and natural language processing (NLP ) techniques. The generated API mashups are further ranked and recommended to developers based on a quantitative indicator of whether the given API mashup is plausible. To evaluate the overall accuracy of the proposed method, we use the generated API mashups to train several machine learning and deep learning models, and then use an independent mashup dataset collected from Github projects for testing. The experimental results show that our proposed method is feasible and accurate for automatic API mashup generation and recommendation.

Image retrieval has become an important function in many emerging computer vision applications e.g. online shopping via images, medical health care systems. More and more images are being generated and stored in public clouds. However, recent photo leakage events raise concerns about privacy leaks for images stored in public clouds. In this paper, we present an efficient scalable hierarchical image retrieval system (CASHEIRS) which provides privacy-aware image retrieval feature. CASHEIRS employs transformed Convolutional Neural Network features to improve image retrieval accuracy and an encrypted hierarchical index tree to speed up the query process. Extensive evaluations using Caltech256 and INRIA Holiday datasets show that CASHEIRS is more effective than three existing schemes. We also demonstrate its practicality on a mobile device.

Recent studies have shown that autistic children tend to speak and interact more with an interactive robot. Unfortunately, due to the high deployment cost, many robotic experiments were still conducted in highly controlled clinical or home settings. Hence, a low cost robot needs to be designed to benefit families with autistic children. In Summer 2014, we have designed a low cost robot, LILI version 1 with limited features e.g. hard coded commands. Since then, we have added improved features to LILI such as adding a natural language processing engine so that LILI can understand more human-like commands. In this paper, we describe two more new features we add to LILI, namely (i) interactive story telling, and (ii) object recognition. We present how we design these two features. We also briefly describe how we re-design the software architecture to make it more modular.

The rapid pace of urbanization and socioeconomic development encourage people to spend more time together and therefore monitoring of human dynamics is of great importance, especially for facilities of elder care and involving multiple activities. Traditional approaches are limited due to their high deployment costs and privacy concerns (e.g., camera-based surveillance or sensor-attachment-based solutions). In this work, we propose to provide a fine-grained comprehensive view of human dynamics using existing WiFi infrastructures often available in many indoor venues. Our approach is low-cost and device-free, which does not require any active human participation. Our system aims to provide smart human dynamics monitoring through participant number estimation, human density estimation and walking speed and direction derivation. A semi-supervised learning approach leveraging the non-linear regression model is developed to significantly reduce training efforts and accommodate different monitoring environments. We further derive participant number and density estimation based on the statistical distribution of Channel State Information (CSI) measurements. In addition, people's walking speed and direction are estimated by using a frequency-based mechanism. Extensive experiments over 12 months demonstrate that our system can perform fine-grained effective human dynamic monitoring with over 90% accuracy in estimating participants number, density, and walking speed and direction at various indoor environments.

A key component of a smart grid is its ability to collect useful information from a power grid for enabling control centers to estimate the current states of the power grid. Such information can be delivered to the control centers via wireless or wired networks. It is envisioned that wireless technology will be widely used for local-area communication subsystems in the smart grid (e.g., in distribution networks). However, various attacks with serious impact can be launched in wireless networks such as channel jamming attacks and denial-of-service attacks. In particular, jamming attacks can cause significant damages to power grids, e.g., delayed delivery of time-critical messages can prevent control centers from properly controlling the outputs of generators to match load demands. In this paper, a communication subsystem with enhanced self-healing capability in the presence of jamming is designed via intelligent local controller switching while integrating a retransmission mechanism. The proposed framework allows sufficient readings from smart meters to be continuously collected by various local controllers to estimate the states of a power grid under various attack scenarios. The jamming probability is also analyzed considering the impact of jammer power and shadowing effects. In addition, guidelines on optimal placement of local controllers to ensure effective switching of smart meters under jamming are provided. Via theoretical, experimental and simulation studies, it is demonstrated that our proposed system is effective in maintaining communications between smart meters and local controllers even when multiple jammers are present in the network.

Affordable cloud computing technologies allow users to efficiently store, and manage their Personal Health Records (PHRs) and share with their caregivers or physicians. This in turn improves the quality of healthcare services, and lower health care cost. However, serious security and privacy concerns emerge because people upload their personal information and PHRs to the public cloud. Data encryption provides privacy protection of medical information but it is challenging to utilize encrypted data. In this paper, we present a privacy-preserving disease treatment, complication prediction scheme (PDTCPS), which allows authorized users to conduct searches for disease diagnosis, personalized treatments, and prediction of potential complications. $PDTCPS$ uses a tree-based structure to boost search efficiency, a wildcard approach to support fuzzy keyword search, and a Bloom-filter to improve search accuracy and storage efficiency. In addition, our design also allows health care providers and the public cloud to collectively generate aggregated training models for disease diagnosis, personalized treatments and complications prediction. Moreover, our design provides query unlinkability and hides both search & access patterns. Finally, our evaluation results using two UCI datasets show that our scheme is more efficient and accurate than two existing schemes.

Existing human action recognition systems for 3D sequences obtained from the depth camera are designed to cope with only one action category, either single-person action or two-person interaction, and are difficult to be extended to scenarios where both action categories co-exist. In this paper, we propose the category-blind human recognition method (CHARM) which can recognize a human action without making assumptions of the action category. In our CHARM approach, we represent a human action (either a single-person action or a two-person interaction) class using a co-occurrence of motion primitives. Subsequently, we classify an action instance based on matching its motion primitive co-occurrence patterns to each class representation. The matching task is formulated as maximum clique problems. We conduct extensive evaluations of CHARM using three datasets for single-person actions, two-person interactions, and their mixtures. Experimental results show that CHARM performs favorably when compared with several state-of-the-art single-person action and two-person interaction based methods without making explicit assumptions of action category.

In many image retrieval systems, re-ranking is an important final step to improve the retrieval accuracy given an initial ranking list. K-Nearest Neighbors (k-NN) re-ranking algorithms are the class of algorithms that re-rank an initial ranked list by comparing the similarity between a query image's k-NN and the k-NN of candidate database images, e.g. the initially high ranked images. In this paper, we present a novel 2-stage k-NN re-ranking algorithm. In stage one, we generate an expanded list of candidate database images for re-ranking so that some lower ranked ground truth images will be included for the next stage. In stage two, we re-rank the list of candidate images using a confidence score which is calculated based on both the ranking consistency and reciprocal k-NN properties. Our experimental results on two popular benchmark datasets along with a large-scale 1 million distraction dataset show improved performance over existing k-NN re-ranking methods.

In recent years, enabling computer systems to recognize facial expressions and infer emotions from them in real time has become very important since such information can be used in emerging applications such as video games, educational software, computer-based tutoring for special need children for better human computer interactions. However, real time emotion recognition using video streams face challenges due to the varying illuminations. In this paper, we present a real time emotion recognition scheme using dense optical flow based approach and SVM classifier. Via extensive analysis using newly collected datasets of 370 videos, we demonstrate that our approach demonstrates high accuracy in recognizing 4 basic emotions: happy, angry, surprise and sad.

Recently, researchers have proposed solutions to build on-device mobile visual search (ODMVS) systems. Different from traditional client-server mobile visual search systems, an ODMVS supports image searching directly within a mobile device. An ODMVS needs to be designed with constrained hardware in mind e.g. limited memory, less powerful CPU. In this paper, we present, EMOD, an efficient on-device mobile visual search system based on the Bag-of-Visual-Word (BOVW) framework but uses a small visual dictionary. An Object Word Ranking (OWR) algorithm is proposed to efficiently identify the most useful visual words of an image so as to construct a compact image signature for fast retrieval and greatly improved retrieval performance. Due to having a small visual dictionary, we propose the Top Inverted Index Ranking scheme to reduce the number of candidate images for similarity calculation. In addition, EMOD adopts a more efficient version of the recently proposed Ranking Consistency re-ranking algorithm for further performance enhancement. Via extensive experimental evaluations, we demonstrate that our prototype EMOD system yields good retrieval accuracy and query response times for a database with over 10K images.

Many organizations and individual users are out-sourcing their information which includes sensitive data into the cloud. To deal with the potential risks of privacy exposure, such data is typically encrypted before being outsourced but users would like to conduct keyword-based searches. Traditional searchable encryption techniques are overly restrictive for they only allow exact keyword search. Thus, fuzzy keyword search is needed to deal with typos in users' search strings. In this paper, we present a Cuckoo Filter Based Private Keyword Search Scheme (CFPKS) to provide privacy-aware keyword search over encrypted data. This CFPKS scheme uses a bed-tree structure-based index to boost search efficiency, a wildcard approach to support fuzzy keyword search, and a Cuckoo-filter to improve search accuracy and storage efficiency. Our scheme handles both typos and query unlinkability. Using a large ACM publication dataset, the evaluation results comparing the search efficiency and accuracy of our proposed CFPKS scheme with three existing schemes show that our scheme achieves higher search accuracy with lower search cost.

Recent advances in visual tracking have focused on handling deformations and occlusions using the part-based appearance model. However, it remains a challenge to come up with a reliable target representation using local parts, and hence existing trackers continue to face drifting problems. To deal with this challenge, we propose a robust online model, formulating the tracking task as a problem of identifying Temporally Coherent Part (TCP) clusters. Specifically, we pose the TCP clusters identification task as a dense neighborhoods searching problem using a relational hyper graph in which the relationship among multiple temporal local parts is encoded as the affinity value of a hyper edge connecting them. Such high-order relations ships among multiple local parts across the temporal domain make our tracker more robust towards deformations and occlusions. Extensive experiments on various challenging video sequences demonstrate that our TCP-based method performs better than the state-of-the-art methods.

The rapid deployment of sensing technology in smartphones and the explosion of their usage in people's daily lives provide users with the ability to collectively sense the world. This leads to a growing trend of mobile healthcare systems utilizing sensing data collected from smartphones with/without additional external sensors to analyze and understand people's physical and mental states. However, such healthcare systems are vulnerable to user spoofing, in which an adversary distributes his registered device to other users such that data collected from these users can be claimed as his own to obtain more healthcare benefits and undermine the successful operation of mobile healthcare systems. Existing mitigation approaches either only rely on a secret PIN number (which can not deal with colluded attacks) or require an explicit user action for verification. In this paper, we propose a user verification system leveraging unique gait patterns derived from acceleration readings to detect possible user spoofing in mobile healthcare systems. Our framework exploits the readily available accelerometers embedded within smartphones for user verification. Specifically, our user spoofing mitigation framework (which consists of three components, namely Step Cycle Identification, Step Cycle Interpolation, and Similarity Comparison) is used to extract gait patterns from run-time accelerometer measurements to perform robust user verification under various walking speeds. We show that our framework can be implemented in two ways: user-centric and server-centric, and it is robust to not only random but also mimic attacks. Our extensive experiments using over 3,000 smartphone-based traces with mobile phones placed on different body positions confirm the effectiveness of the proposed framework with users walking at various speeds. This strongly indicates the feasibility of using smartphone based low grade accelerometer to conduct gait recognition and facilitate effective user verification without active user cooperation.

This paper investigates the routing protocol in smart phone-based mobile Ad-Hoc networks. We introduce a new dual radio communication model, where a long-range, low cost, and low rate radio is integrated into smart phones to assist regular radio interfaces such as WiFi and Bluetooth. We propose to use the long-range radio to carry out small management data packets to improve the routing protocols. Specifically, we develop new schemes built on the long-range radio to improve the efficiency of the path establishment process in the existing on-demand Ad-Hoc routing protocols. We have prototyped our solution LAAR on Android phones and evaluated the performance with small scale experiments and large scale simulation implemented on NS2. The results show that LAAR significantly improve the performance in terms of the overhead and the number of messages transferred in the network.

Recent studies show that autistic children tend to speak and interact more in the presence of an interactive robot. Unfortunately, most of the robotic experiments were conducted in highly controlled clinical settings or limited selected home environments due to the high deployment cost. In this paper, we design a low-cost interactive robot that can be readily deployed to home environments. Our robot, called Lehigh Instrument for Learning Interaction (LILI), interacts with users via gestures, voice commands, and an animated speaking avatar. LILI can recognize users' faces, and her motion can be controlled either via gesture or voice. We describe both the system and software architectures of LILI. Experimental results are also presented.

Mobile wireless devices play important roles in our daily life, e.g., users often use such devices to take pictures and share with friends via opportunistic peer-to-peer links, which however are intermittent in nature, and hence require the store-and-forward feature proposed in Delay Tolerant Networks to provide useful data sharing opportunities. Moreover, mobile devices may not be willing to forward data items to other devices due to the limited resources. Hence, effective data dissemination schemes need to be designed to encourage nodes to collaboratively share data. We propose a Multi-Receiver Incentive-Based Dissemination (MuRIS) scheme that allows nodes to cooperatively deliver information of interest to one another via chosen paths utilizing few transmissions. Our scheme exploits local historical paths and users' interests information maintained by each node. In addition, the charge and rewarding functions incorporated within our scheme stimulate cooperation among nodes such that the nodes have no incentive to launch edge insertion attacks. Furthermore, our charge and rewarding functions are designed such that the chosen delivery paths mimic efficient multicast tree that results in fewer delivery hops. Extensive simulation studies using real human contact-based mobility traces show that our scheme outperforms existing methods in terms of delivery ratio and transmission efficiency.

Content centric network (CCN) enables users to retrieve contents of interest(e.g. videos, music etc.) using the content names directly. Researchers face two critical challenges in CCN design: content naming and mobility support. The content naming is a fundamental building block for CCN. Users must be able to obtain names of contents of interest fast so that contents can be correctly retrieved. Meanwhile, the rapid development of wireless technologies especially the emergence of cellular 4G networks allows users to retrieve content on the move. However, since mobile users may frequently change their attachment points (e.g. 4G, WiFi, WiMax etc.), the CCN architecture must be able to handle mobility effectively so that a user can still get the content when being connected to a different attachment point. In this paper, we propose SCOM, a scalable content centric network architecture with mobility support, to solve those two issues. SCOM adopts an improved keyword-based naming resolution method to enable faster name retrieval. Users could retrieve names of contents of interest by sending keyword queries to the corresponding Content Resolution Servers (CRS). For mobility, SCOM provides efficient solutions for both keyword based queries, and content retrievals for users on the move. Via large scale simulation studies using real internet topology, we demonstrated that our solutions are effective in terms of shorter query response times and faster content retrievals during mobility.

Traditionally, content-based image retrieval systems (CBIR) are designed to allow users to search for images in large databases which match closely with users' query images. Recent emergence of powerful mobile devices equipped with digital cameras have led to the emergence of several interesting mobile CBIR applications. Due to the limited resources in mobile devices, it is critical that the image matching engine within any mobile CBIR system be efficiently designed. Many existing image matching engines use SURF-based methods which return many key points, and hence are not quite suitable for mobile devices. In this paper, we present an efficient mobile visual search system (EMOVIS) which allows mobile users to retrieve relevant information using image-based queries. EMOVIS uses two unique salient key point identification schemes we designed which allow image matching to be conducted efficiently and with high accuracy. In addition, EMOVIS includes an image cropping scheme which eliminates irrelevant regions within a query image. Such cropping minimizes query latency, bandwidth usage and the energy cost of using EMOVIS. Via extensive evaluations using ZuBuD dataset and our own image dataset, we showed that EMOVIS can achieve higher than 92% accuracy with low computational and energy cost.

With the availability of powerful mobile devices and cellular or WiFi networks with larger bandwidth, users can search and retrieve contents everywhere anytime. Recently content-centric networks have been proposed to provide users with more flexible access to contents than the existing IP-based networks. We have designed secure content centric mobile network (SECON)s that allow users to publish and retrieve contents securely. Unique SECON features include supporting keyword-based content queries, and enhanced attribute-based security approach. In this paper, we describe two new features for SECON, namely real-time video streaming and mobility support. We further describe Lehigh Explorer, an Android application we developed for users to explore different campuses in real-time or virtually. Users of Lehigh Explorer issue keyword based content queries to retrieve data items of interests. They can also retrieve streaming videos of interests while on the move. Our prototype evaluation using GENI test bed showed that the average handoff time for our streaming video service is 732 ms without caching and 101 ms with caching. The energy consumption for supporting video streaming of different qualities using 3 Android-based phones is also reported.

The rapid deployment of sensing technology in smartphones and the explosion of their usage in people's daily lives provide users with the ability to collectively sense the world. This leads to a growing trend of mobile healthcare systems utilizing sensing data collected from smartphones with/without additional external sensors to analyze and understand people's physical and mental states. However, such healthcare systems are vulnerable to user spoofing attacks, in which an adversary distributes his registered device to other users such that data collected from these users can be claimed as his own to obtain more healthcare benefits and undermine the successful operation of mobile healthcare systems. Existing mitigation approaches either only rely on a secret PIN number (which can not deal with colluded attacks) or require an explicit user action for verification. In this paper, we propose a user verification scheme leveraging unique gait patterns derived from acceleration readings in mobile healthcare systems to detect possible user spoofing attacks. Our framework exploits the readily available accelerometers embedded within smartphones for user verification. Specifically, our user spoofing attack mitigation scheme (which consists of three components, namely Step Cycle Identification, Step Cycle Interpolation, and Similarity Score Computation) is used to extract gait patterns from run-time accelerometer measurements to perform robust user verification under various walking speeds. Our experiments using 322 smartphone-based traces over a period of 6 months confirm that our scheme is highly effective for detecting user spoofing attacks. This strongly indicates the feasibility of using smartphone based low grade accelerometer to conduct gait recognition and facilitate effective user verification without active user cooperation.

The electric power grid is a crucial infrastructure in our society and is always a target of malicious users and attackers. In this paper, we first introduce the concept of unidentifiable attack, in which the control center cannot identify the attack even though it detects its presence. Thus, the control center cannot obtain deterministic state estimates, since there may have several feasible cases and the control center cannot simply favor one over the others. Given an unidentifiable attack, we present algorithms to enumerate all feasible cases, and propose an optimization strategy from the perspective of the control center to deal with an unidentifiable attack. Furthermore, we propose a heuristic algorithm from the view of an attacker to find good attack regions such that the number of meters required to compromise is as few as possible. We also formulate the problem that how to distinguish all feasible cases if the control center has some limited resources to verify some meters, and solve it with standard algorithms. Finally, we briefly evaluate and validate our enumerating algorithms and optimization strategy.

Mobile wireless devices (e.g., smartphones, PDAs, and notebooks) play important roles in our daily life, e.g., users often use such devices for bank transactions, keep in touch with friends. Users can also store such information and share with one another via opportunistic peer to peer links. However, peer to peer links are opportunistic links which are intermittent in nature and hence require the store-and-forward feature proposed in Delay Tolerant Networks to provide useful data sharing opportunities. Moreover, due to the limited resources, e.g., communication bandwidth and battery consumption, mobile devices can be selfish and may not be willing to forward data items to other devices that are interested in such items. Hence, effective data dissemination schemes need to be designed to encourage nodes to collaboratively share data. In this paper, we propose a Multi-Receiver Incentive-Based Dissemination (MuRIS) scheme that allows nodes to cooperatively deliver information of interest to one another via chosen delivery paths that utilize few transmissions. Our MuRIS scheme utilizes local historical path and tracks users' interests information maintained by each node. In addition, the charge and reward functions incorporated within our MuRIS scheme stimulate cooperation among nodes such that the nodes have no incentive to launch edge insertion attacks. Furthermore, our charge and reward functions are designed such that the chosen delivery paths mimic efficient multicast tree that results in fewest delivery hops. Extensive simulation studies using real human contact-based mobility traces show that our MuRIS scheme outperforms existing methods in terms of delivery ratio and transmission efficiency.

The inclusion of embedded sensors in mobile phones, and the explosion of their usage in people's daily lives provide users with the ability to collectively sense the world. The collected sensing data from such a mobile phone enabled social network can be mined for users' behaviors and their social communities, and to support a broad range of applications including mobile healthcare systems. However, such mobile healthcare systems built upon social networks are vulnerable to clone attacks, in which the adversary replicates the legitimate nodes and distributes the clones throughout the network to undermine the successful application deployment. Existing clone attack mitigation approaches either only focus on the prevention techniques or can only work in static or well-connected networks, and hence are not applicable to our targeted mobile healthcare systems. In this paper, we propose a social closeness based method in a mobile healthcare disease control system to detect any clone attacks that may be launched to disrupt the normal operations of the system. Our social closeness based method exploits the social relationships among users for clone attack detection. Specifically, we define a new metric called community betweenness, which considers mobile users' community information. We find that the value of this metric changes significantly under the clone attack, which is suitable to be used for clone attack detection. We derive both analytical and training based approaches to determine the threshold setting of the community betweenness for robust clone attack detection. Extensive trace-driven simulation studies reveal that our social closeness based method can detect clone attacks with high detection ratio and low false positive rate.

The emergence of powerful mobile devices has allowed users to publish more contents in the Internet in recent years. The existing Internet architecture cannot cope with such exponential growth in users published contents. Content-centric networks have been proposed recently to allow future Internet to be data-centric rather than network centric. Several content centric networking approaches have been proposed, but most of them assume that users know the unique identifiers of the contents that are of interests to them. SECON [1] proposed a content centric mobile network solution that provides keyword based retrievals. However, the authors do not provide detailed description on how their solution can be made scalable. In this paper, we propose two scalable solutions for keyword based retrievals in content centric networks. Our preliminary simulation results indicate that our solutions are scalable.

Rapid advancements of wireless technologies allow users to access real-time data, and stay connected with friends and business while they are on the move. However, most emerging mobile applications assume users have cellular data services, and hence not everyone can enjoy new mobile applications. In addition, some emerging mobile applications e.g. mobile recommender system are data-centric but existing IP oriented communication paradigms are not flexible enough to support such applications. In this paper, we present a new secure content centric mobile network that supports content centric communication paradigm. Users can exchange information using peer to peer mode without having to rely on cellular data services. Content-centric security solution is provided where data owners can share encrypted published data items with others without knowing a priori who the interested users may be. Our preliminary prototype deployed in the ORBIT testbed demonstrates some of the key features we have designed.

Community association is an important attribute of a social network because people may belong to varying groups with different characteristics at different times. Traditional community detection approaches often rely on a centralized server and are only useful for offline data analysis. In this paper, we propose and evaluate a distributed community detection approach that allows individual users to detect their own communities based on local observations. Our proposed template- matching method derives dynamic spatial and temporal characteristics of social communities by exploiting human's mobility patterns. Our template matching method allows users with similar moving patterns to be grouped together as one community. Our results using both simulation as well as real experiments demonstrate that our method can detect local communities effectively with high detection rate and low false positive rate.

New mobile phones equipped with multiple sensors provide users with the ability to sense the world at a microscopic level. The collected mobile sensing data can be comprehensive enough to be mined not only for the understanding of human behaviors but also for supporting multiple applications ranging from monitoring/tracking, to medical, emergency and military applications. In this work, we investigate the feasibility and effectiveness of using human contact traces collected from mobile phones to derive social community information to control the disease propagation rate in the healthcare domain. Specifically, we design a community-based framework that extracts the dynamic social community information from human contact based traces to make decisions on who will receive disease alert messages and take vaccination. We have experimentally evaluated our framework via a trace-driven approach by using data sets collected from mobile phones. The results confirmed that our approach of utilizing mobile phone enabled dynamic community information is more effective than existing methods, without utilizing social community information or merely using static community information, at reducing the propagation rate of an infectious disease. This strongly indicates the feasibility of exploiting the social community information derived from mobile sensing data for supporting healthcare related applications.

Powerful wireless devices carried by humans can form human contact-based networks. Such networks often suffer from intermittent connectivity. Thus, providing an effective information dissemination feature in such networks is very important. In this paper, we explore a cooperative user centric information dissemination scheme which allows published data items to be delivered to interested nodes efficiently. Our scheme uses fewer relays and allows each node to operate distributedly using locally gathered information. Our scheme is more effective than the epidemic scheme since it achieves comparable success ratio with a 45-60% reduction in storage requirement and 47-53% reduction in transmissions. We also compare our scheme with an ideal scheme which assumes one can analyze contact traces apriori to determine their dominating sets, and show that our scheme can be more efficient than this ideal scheme.

Recently, multichannel protocols have been proposed and shown to significantly improve the aggregate throughput compared to single channel protocols. However, some existing proposals still suffer from low channel utilization or unfair bandwidth allocation between different flows in a multihop wireless network. In this paper, we design an asynchronous multichannel MAC called the cooperative multichannel MAC (COMMAC) that allows nodes to identify useful data channels using the channel usage information collected by their neighbors. Channel usage information includes primary users' activities. It also utilizes a new control message to prevent sending nodes from doing unnecessary backoffs. Via simulations, we compare our designed COMMAC with the AMCP scheme and show that our scheme can achieve 35% more throughput in multihop adhoc networks.

The Disruption Tolerant Networks (DTNs) are vulnerable to insider attacks, in which the legitimate nodes are compromised and the adversary modifies the delivery metrics of the node to launch harmful attacks in the networks. The traditional detection approaches of secure routing protocols can not address such kind of insider attacks in DTNs. In this paper, we propose a mutual correlation detection scheme (MUTON) for addressing these insider attacks. MUTON takes into consideration of the transitive property when calculating the packet delivery probability of each node and correlates the information collected from other nodes. We evaluated our approach through extensive simulations using both Random Way Point and Zebranet mobility models. Our results show that MUTON can detect insider attacks efficiently with high detection rate and low false positive rate.

The Disruption Tolerant Networks (DTNs) are especially useful in providing mission critical services such as in emergency networks or battlefield scenarios. However, DTNs are vulnerable to insider attacks, in which the legitimate nodes are compromised and the adversary nodes launch blackhole attacks by dropping packets in the networks. The traditional approaches of securing routing protocols can not address such insider attacks in DTNs. In this paper, we propose a method to secure the history records of packet delivery information at each contact so that other nodes can detect insider attacks by analyzing these packet delivery records. We evaluated our approach through extensive simulations using both Random Way Point and Zebranet mobility models. Our results show that our method can detect insider attacks efficiently with high detection rate and low false positive rate.

Some ad hoc network scenarios are characterized by frequent partitions and intermittent connectivity. A store-and-forward network architecture known as the disruption tolerant network (DTN) has been designed for such challenging network environments. To further improve the delivery performance, some researchers have proposed some network coding schemes for DTNs. However, not much papers discuss the security issues of network coding schemes in DTNs. In this paper, we first discuss some attacks that can be launched against network coding schemes in DTNs. Then, we focus on evaluating the impact of selective data dropping attacks on the delivery performance of a network coding scheme we design for DTN. Next, we describe a mitigation scheme that we design to overcome such attacks. Our mitigation scheme uses dynamic redundancy factor to generate more coded packets when a source notices performance degradation in the delivery performance. Via simulation studies, we show that our mitigation scheme is effective in restoring the performance degradation caused by the selective dropping attacks as long as alternate DTN paths exist for a source/destination pair.

Some ad hoc network scenarios are characterized by frequent partitions and intermittent connectivity. Hence, existing adhoc routing schemes which assume the existence of end-to-end paths do not work in such challenging networks. Disruption tolerant networking (DTN) technology has been designed for such challenging network environments. Several unicast and multicast routing schemes have been designed for DTNs. However, the existing multicast routing schemes assume a route discovery process that is similar to the existing adhoc network routing approach, and hence will not work well in very sparse network scenarios. Thus, in this paper, we explore an encounter-based multicast routing (EBMR) scheme for DTNs. Our scheme uses fewer hops for message delivery. We present an analytical framework for estimating the delivery performance of the EBMR scheme, and present some analytical and simulation results to show that the EBMR scheme can achieve higher delivery ratio while maintaining high data transmission efficiency compared to other multicast strategies.

Mobile nodes in some challenging network scenarios suffer from intermittent connectivity and frequent partitions e.g. battlefield and disaster recovery scenarios. Disruption tolerant network (DTN) technologies are designed to enable nodes in such environments to communicate with one another. In the past, we have proposed two intradomain multicast routing schemes, namely the context-aware multicast routing (CAMR) and encounter-based multicast routing (EBMR) schemes. In this paper, we consider the problem of routing multicast messages across different domains. We present a ferry based interdomain multicast delivery scheme where a ferry is used to deliver multicast messages across groups that are partitioned and a variant of the encounter-based multicast routing scheme is used as the intradomain routing scheme for intradomain delivery. We then present simulation results using different group mobility models to illustrate the usefulness of the scheme we design. Our results indicate that the scheme we design can achieve high delivery ratio with reasonable data efficiency. Our results also indicate that the delivery ratio seen using a more realistic VANET model is slightly worse than that seen using the RPGM model.

Designing a routing scheme for partitioned ad-hoc networks is challenging since end-to-end connectivity may not exist. A Message ferrying (MF) scheme has been proposed recently for delivering non-real-time traffic. In this ferrying scheme, message ferries are used to collect and deliver packets. In the past, we have designed an elliptical zone forwarding (EZF) scheme for a ferry to deliver messages among partitioned nodes that are moving around. In this paper, we first discuss a weakness of the EZF scheme. Then, we describe three ferry route design with lookahead schemes that are capable of achieving higher delivery ratio, namely (a) the Minimum Weighted Sum First (MWSF) scheme, (b) the fixed K-lookahead scheme (FKLAS) and (c) the dynamic lookahead scheme (DLAS). We also present extensive simulation results that compare the different schemes in various scenarios e.g. with different node mobility models, message deadlines etc. Our results indicate DLAS provides the best delivery performance, followed by FKLAS and the MWSF scheme. I.

Disruption Tolerant Network (DTN) technologies are emerging solutions to networks that experience frequent partitions. In this paper, we propose the scalable hierarchical inter-domain multicast (SHIM) approach for DTNs. SHIM has the following characteristics: i) it is capable of delivering multicast messages to receivers distributed in different domains; ii) the size of the membership information maintained by the source leader is determined by its out-degree in the leader layer, no matter how large the number of the real receivers is; and iii) it at least doubles the message delivery efficiency than that of directly extending the existing intra-domain DTN multicast methods to perform the inter-domain multicast operations. Our results also show that the message delivery ratio of SHIM can be improved to be almost 100% when the custodian transfer functionality is enabled in the overall networks.

Several intrusion detection approaches have been proposed for mobile ad hoc networks. Many of the approaches assume that there are sufficient neighbors to help monitor the transmissions and receptions of data packets by other nodes to detect abnormality. However, in a sparsely connected adhoc network, nodes usually have very small number of neighbors. In addition, new history based routing schemes e.g. Prophet have been proposed because traditional adhoc routing schemes do not work well in sparse ad hoc networks. In this paper, we propose a ferry-based intrusion detection and mitigation (FBIDM) scheme for sparsely connected ad hoc networks that use Prophet as their routing scheme. Via simulations, we study the effectiveness of the FBIDM scheme when malicious nodes launch selective data dropping attacks. Our results with different mobility models, ferry speed, traffic load scenarios indicate that the FBIDM scheme is promising in reducing the impact of such malicious attacks.

Disruption Tolerant Network (DTN) is characterized by frequent partitions and intermittent connectivity. Power management issue in such networks is challenging. Existing power management schemes for wireless networks cannot be directly applied to DTNs because they assume the networks are well-connected. Since the network connectivity opportunities are rare, any power management scheme deployed in DTNs should not worsen the existing network connectivity. In this paper, we design a power management scheme called context-aware power management scheme (CAPM) for DTNs. Our CAPM scheme has an adaptive on period feature that allows it to achieve high delivery ratio and low delivery latency when used with Prophet, a recently proposed DTN routing scheme. Via simulations, we evaluate the performance of the CAPM scheme when used with the Prophet routing scheme in different scenarios e.g. different traffic load, node speeds and sleep patterns. Our evaluation results indicate that the CAPM scheme is very promising in providing energy saving (as high as 80%) without degrading much the data delivery performance.

Anomalous interdomain border gateway protocol (BGP) events including misconfigurations, attacks and large-scale power failures often affect the global routing infrastructure. Thus, the ability to detect and categorize such events is extremely useful. In this article we present a novel anomaly detection technique for BGP that distinguishes between different anomalies in BGP traffic. This technique is termed higher order path analysis (HOPA) and focuses on the discovery of patterns in higher order paths in supervised learning datasets. Our results demonstrate that not only worm events but also different types of worms as well as blackout events are cleanly separable and can be classified in real time based on our incremental approach. This novel approach to supervised learning has potential applications in cybersecurity/forensics and text/data mining in general.

Malicious network traffic, including widespread worm activity, is a growing threat to Internet-connected networks and hosts. In this paper, we propose a network intrusion detection system (NIDS) with semantics-aware capability. Our NIDS segregates suspicious traffic from the regular traffic flow, extracts binary code from the suspicious traffic, and performs semantic analysis on it to identify potential threats. Our contributions in this work are threefold: (a) we believe our prototype is the first NIDS that provides semantics-aware capability, (b) our implementation is more efficient than what is reported in (M. Christodorescu et al., 2005) (c) our designed templates can capture polymorphic shellcodes with added sequences of stack and mathematic operations.

Previous studies have shown that current inter-domain routing protocol, border gateway protocol (BGP), is vulnerable to various attacks. Initially, the major concern about BGP security is that malicious BGP routers can arbitrarily falsify BGP routing messages and spread incorrect routing information. Recently, some authors have pointed out the impact of a type of attack, namely selective dropping attack that has not studied before. The authors have shown that such an attack can result in data traffic being blackholed or trapped in a loop. However, the authors did not elaborate on how one can detect selective dropping attacks. In this paper, we present a scheme we designed to detect selective dropping attacks in BGP. We conducted extensive experiments in DETER to evaluate the effectiveness of our scheme using three 30-node AS topologies generated from Brite. Our study shows that our scheme is quite promising

Delay and disruption tolerant networks have been proposed to address data communication challenges in network scenarios where an instantaneous end-to-end path between a source and destination may not exist, and the links between nodes may be opportunistic, predictably connectable, or periodically-(dis)connected. In this paper, we describe the store-and-forward and custody transfer concepts that are used in DTNs. Then, we present simulation results that illustrate the usefulness of the custody transfer feature, and a message ferry in improving the end-to-end message delivery ratio in a multihop scenario where link availability can be as low as 20%. In particular, our results indicate that one can achieve a delivery ratio as high as 90-99% with appropriate buffer allocations. We also provide some preliminary insights on the design factors that influence the end to end delivery ratio, e.g., the link availability patterns and buffer allocation strategies

Disruption tolerant networks (DTNs) are emerging solutions to networks that experience frequent network partitions and large end-to-end delays. In this paper, we study how to provide high-performance multicasting service in DTNs. We develop a multicasting mechanism based on on-demand path discovery and overall situation awareness of link availability (OS-multicast) to address the challenges of opportunistic link connectivities in DTNs. Simulation results show that OS-multicast can achieve a better message delivery ratio than existing approaches, e.g. DTBR (a dynamic tree-based routing), with similar delay performance. OS-multicast also achieves better efficiency performance when the probability of link unavailability is high and the duration of link downtime is large

Standard ad hoc routing protocols do not work in intermittently connected networks since end-to-end paths may not exist in such networks. A store-and-forward approach (K. Fall, 2003) has been proposed for such networks. The nodes in such networks move around. Thus, the proposed delay tolerant network (DTN) architecture (K. Fall, 2003) needs to be enhanced with a mobility management scheme to ensure that nodes that wish to correspond with mobile hosts have a way of determining their whereabouts. The mobile hosts may move a short distance and hence remain within the vicinity of a DTN name registrar (DNR) (one communication link away) or they may move far away (multiple communication links away). In this paper, we present the mobility management scheme we propose for DTN environments. In addition, we provide simple analytical formulae to evaluate the latency required for performing location updates, and the useful utilization that each node can use for data transfer assuming that the communication links between nodes are periodically available for a short period of time. Our simple analytical model allows us to draw insights into the impact of near/far movements on the useful utilization

Distributed denial-of-service (DDoS) attacks are a critical threat to the Internet. This paper introduces a DDoS defense scheme that supports automated online attack characterizations and accurate attack packet discarding based on statistical processing. The key idea is to prioritize a packet based on a score which estimates its legitimacy given the attribute values it carries. Once the score of a packet is computed, this scheme performs score-based selective packet discarding where the dropping threshold is dynamically adjusted based on the score distribution of recent incoming packets and the current level of system overload. This paper describes the design and evaluation of automated attack characterizations, selective packet discarding, and an overload control process. Special considerations are made to ensure that the scheme is amenable to high-speed hardware implementation through scorebook generation and pipeline processing. A simulation study indicates that packetscore is very effective in blocking several different attack types under many different conditions.

Data communication challenges exist in some emerging network scenarios where an instantaneous end-to-end path between a source and destination may not exist, and the links between nodes may be opportunistic, predictably connectable, or periodically-(dis)connected. We propose an enhanced disruption tolerant network architecture to address such challenges. In this paper, we present a generalized naming convention for the enhanced DTN architecture that permits separate representations based on network topology, administrative control, physical location, and other factors. In addition, we illustrate possible system operations in this enhanced DTN architecture such as DTN neighbor discovery, gateway selection, mobility management, and route discovery.

The paper presents a performance study of two algorithms for dynamic allocation of the dedicated channels (DCH) in UMTS over the air interface, namely least recently used (LRU) algorithm and an adaptive algorithm. The algorithms are designed to efficiently share the dedicated channels among users whose traffic patterns are characterized by bursty packet transfers followed by long inactivity periods. Specifically, Web browsing, FTP and email applications were considered in order to evaluate the performance of the above mentioned resource allocation algorithms in the context of bursty traffic with relaxed delay constraints and In the presence of delays introduced by the backhaul network

We propose three inter-domain routing schemes for ad hoc networks, namely the implicit foreign degree based protocol (IFD), the explicit locally optimal protocol (ELO) and the explicit limited scope protocol (ELS). We studied the performance of these different schemes. Our simulation studies reveal that the explicit schemes can achieve high packet delivery ratio and good average end-to-end delay at a lower overhead cost compared to IFD. We also compared our approaches with LANMAR, an existing scalable routing protocol for ad hoc networks. The results show that our approach outperforms LANMAR in terms of the packet delivery ratio and control overhead, while maintaining comparable end-to-end delay at low and medium mobility rate (less than 8 m/s).

Message ferrying (MF) (Wenrui Zhao and Ammar, M.H., Proc. IEEE Workshop on Future Trends in Distrib. Computing Syst., 2003), a viable solution for routing in highly partitioned ad-hoc networks, exploits message ferries to transfer packets between disconnected nodes. The paper studies the delivery quality of service (QoS) for certain urgent messages in the constrained and the relaxed constrained MF systems. Efficient algorithms to compute near-optimal ferry routes are proposed, delay analysis is conducted and the results are compared to the non-constrained scenario.

In this paper, we first describe the UMTS architecture for supporting MBMS. Then, we describe various airlink options for carrying multicast traffic in UMTS Release 99/4 system. One can carry it over FACH or DSCH channel. We discuss why DSCH channel is not an attractive solution. Then, we discuss how rate splitting, longer TTI and STTD techniques can be used to reduce the power requirement of delivering multicast traffic over the FACH channel for MBMS users. Another possibility is to use a combination of multicast as well as dedicated channels to serve all MBMS users. Some users who are in bad conditions are served using dedicated channels and those who are in relatively good conditions are served using a multicast channel. Last but not least, we present simulation results to give us an idea of how much power is required to support MBMS.

Distributed denial of service (DDoS) attack is a critical threat to the Internet. Recently we have proposed the PacketScore scheme, a DDoS defense architecture that supports automated attack detection, on-line attack characterization and attack blocking. Its key idea is to use a statistics-based packet scoring mechanism to distinguish between legitimate and non-legitimate packets and discard packets based on the packet scores. In order for such an approach to work, we need to perform on-line traffic characterizations, and compare such characterizations with the nominal profiles (generated from past history or off-line analysis). The threshold used for the score-based selective packet discard decision is dynamically adjusted based on the score distribution of recent incoming packets. In our previous paper [Kim et al. 2004], we discuss how our proposed system performs in different attack scenarios. In this paper, we first give a brief review of the PacketScore approach and further elaborate on the transient performance under varying attack types and intensities, which may be exploited in more sophisticated attacks. We then show that PacketScore is well capable of blocking such sophisticated attacks by simply adjusting the measurement window time scale to closely track the attack profile.

Distributed denial of service (DDoS) attack is a critical threat to the Internet. Currently, most ISPs merely rely on manual detection of DDoS attacks after which offline fine-grain traffic analysis is performed and new filtering rules are installed manually to the routers. The need of human intervention results in poor response time and fails to protect the victim before severe damages are realized. The expressiveness of existing filtering rules is also too limited and rigid when compared to the ever-evolving characteristics of the attacking packets. Recently, we have proposed a DDoS defense architecture that supports distributed detection and automated on-line attack characterization. We focus on the design and evaluation of the automated attack characterization, selective packet discarding and overload control portion of the proposed architecture. Our key idea is to prioritize packets based on a per-packet score which estimates the legitimacy of a packet given the attribute values it carries. Special considerations are made to ensure that the scheme is amenable to high-speed hardware implementation. Once the score of a packet is computed, we perform score-based selective packet discarding where the dropping threshold is dynamically adjusted based on (1) the score distribution of recent incoming packets and (2) the current level of overload of the system.

Universal Mobile Telecommunications System (UMTS) networks are capable of serving packet-switched data applications at bit rates as high as 384 kbps. This paper studies the capacity and utilization of the downlink of the Iub interface, which lies between the radio network controller (RNC) and the base station (NodeB) in the UMTS network. The 3GPP standards define a Node B "receive window" within which a frame should arrive for it to be processed and transmitted to the UE in time. If the frame arrives too late, it will be discarded. Such frame discard event results in some loss in voice/data quality. Via simulations, we evaluate the link capacity for web-browsing traffic at 64 kbps, 128 kbps and 384 kbps, with a frame discard probability target of 0.5%. Our results indicate that the Iub link utilization is very poor due to the highly bursty nature of data traffic. In order to alleviate this problem, we introduce a rate control (RC) scheme where the peak user data rate is temporarily lowered during times of high congestion. This lowering of data rate is done through appropriate selection of the transport block size within the transport format set. As a result of such rate control, the capacity of the Iub link improves.

Universal Mobile Telecommunication System (UMTS) networks are capable of serving packet-switched data applications at bit rates as high as 384 kbps. Multiplexing multiple data sessions in the radio access networks (RAN) of a UMTS network brings a tremendous capacity gain with respect to the circuit-switched networks, however the utilization of resources are very low due to the bursty nature of data traffic. This paper studies the capacity and utilization of the downlink of the Iub interface, which lies between the radio network controller (RNC) and the base station (NodeB) in a UMTS network. In earlier work, we presented simulation results that indicate that the Iub link utilization is very low with bursty web browsing data especially if each data user is allowed to peak at 384 kbps. Thus, we introduced a rate control algorithm by which the MAC layer constrains the peak rate of a user based on current load conditions. In this paper, we extend that study to the case where multiple service classes (64, 128 and 384 kbps) are multiplexed on the same Iub link. There are some fundamental differences between the ways it is implemented in the multi-service case. We tested the multiservice rate control algorithm with some mixed service scenarios where we noted significant increase in capacity.

UTRAN is a 3G radio access network based on the wide-band DS-CDMA technology. An important feature of UTRAN is its multimedia service capability. UMTS supports high bit rate and variable bit-rate services, packet data and Internet access. When a user sets up a data session, a downlink code is allocated. In UMTS, unless a secondary scrambling code is used, there is a shortage of downlink codes. We study the impacts of inactivity timer values on UMTS system capacity.

We present a first-order end-to-end delay analysis of voice over IP traffic using the UMTS packet bearer service for streaming data in a MS-to-PSTN configuration scenario. Delay components are identified with and without soft handoff support.

UMTS is a third generation radio access network which supports multimedia-capable mobile communication such as packet data and Internet access. In order to provide end-to-end quality of service (QoS) in the UMTS network, certain features need to be incorporated into the MAC design. One possible way of providing different QoS is via providing priority mechanisms. The priority mechanism can be implemented in terms of access priority, service priority and/or buffer management schemes. In this paper, we three access priority schemes based on the random access channel structure. The three schemes are: random chip delay access (RCDAP), random backoff based access (RBBAP), and variable logical channel based access priority (VLCAP). We discuss scenarios where such access priority schemes may be useful. Via simulations, we evaluated and compared these three access priority schemes, Our results indicate that all three schemes provide delay differentiation. However, only the RBBAP scheme provides delay differentiation among different priority classes over a wide range of access request rates without impacting the throughput of the access requests. The other two schemes may cause the throughput of access requests from the lower priority classes to drop apart from forcing them to have a higher access delay.

A study is made of the average delay and the blocking probability of a regulator with a finite buffer size fed with Poisson and batch Poisson streams. The blocking probability depends only on sigma and B through sigma +B where B is the buffer size and sigma is expressed as an integer multiple of the packet size. The provision of regulator buffers allows flexibility in the traffic control such that incoming traffic can be delayed during minor overloads and rejected during major overloads. In addition, the provision of regulator buffers allows the use of smaller network buffers, which are typically more expensive due to high-speed operation. Also studied is the packet departure process from the regulator. A four-parameter descriptor ( sigma , rho , C/sub a//sup 2/, lambda ) is proposed to characterize the output process of the regulator. In addition, formulas are proposed to construct the four-parameter descriptor for the output process of the multiplexer. Such descriptors may be useful in multihop situations. Simulation results indicate that the proposed approximations are close to the simulated values.>