The Job-Shop Scheduling Problem (JSSP), under various forms of manufacturing uncertainty, has recently attracted considerable research attention. Most existing studies focus on parameter uncertainty, such as variable processing times, and typically adopt the actor-critic framework. In this paper, we explore a different but prevalent form of uncertainty in JSSP: structural uncertainty. Structural uncertainty arises when a job may follow one of several routing paths, and the selection is determined not by policy, but by situational factors (e.g., the quality of intermediate products) that cannot be known in advance. Existing methods struggle to address this challenge due to incorrect credit assignment: a high-quality action may be unfairly penalized if it is followed by a time-consuming path. To address this problem, we propose a novel method named UP-AAC. In contrast to conventional actor-critic methods, UP-AAC employs an asymmetric architecture. While its actor receives a standard stochastic state, the critic is crucially provided with a deterministic state reconstructed in hindsight. This design allows the critic to learn a more accurate value function, which in turn provides a lower-variance policy gradient to the actor, leading to more stable learning. In addition, we design an attention-based Uncertainty Perception Model (UPM) to enhance the actor's scheduling decisions. Extensive experiments demonstrate that our method outperforms existing approaches in reducing makespan on benchmark instances.

Dense video captioning (DVC) aims to describe multiple events within a video, and its performance is greatly affected by the accuracy of video event detection. Video event detection involves predicting the proposal boundaries (start and end times) and the classification score of each event in a video. Recently, a few methods have applied diffusion models originally designed for image object detection to detect events in DVC. These methods add noise to the ground-truth event proposal boundaries, and subsequently learn the denoising process. However, these methods often overlook the fundamental differences between videos and images. We observe that, whereas in images the important information for object classification is normally around the boundaries of the ground-truth boxes, in videos the key information for event classification is typically centered in the middle of ground-truth event proposals. As a result, the classification module in these existing diffusion models becomes insensitive to boundary changes introduced by the added noise, leading to sub-optimal performance. This paper introduces DiffDVC, an innovative diffusion model for DVC. The core of DiffDVC is a boundary-sensitive detector. The detector increases the sensitivity of the classification module to boundary changes by focusing on frames within a specific range around the start and end times of noisy event proposals. Additionally, this range is dynamically adjusted to suit different event proposals. Comprehensive experiments on ActivityNet-1.3, ActivityNet Captions, and YouCook2 datasets show DiffDVC achieving superior performance.

In packet erasure networks, federated learning (FL) typically suffers more prohibitive communication overhead from massive retransmissions of high-dimensional gradients. As a result, recent studies are dedicated to developing retransmission-free gradient compression techniques with erasure resilience. Nonetheless, two limitations remain unsolved: existing works neither explore why packet erasure degrades the performance of FL nor exploit the spatial correlations among gradient entries for better compression. In this paper, we investigate FL performance degradation via analyzing model updating deviation and find that the deviation is exacerbated by dependencies among lost gradient entries. On top of this observation, we propose FedRIVQ, a communication-efficient and robust FL framework taking a customized compressor termed random interleaved vector quantization (VQ). FedRIVQ leverages the spatial correlations among gradient entries with VQ and randomly interleaves these entries prior to VQ to eliminate their dependencies. These innovations allow all gradient entries to share an identical erasure probability, thereby packet erasure is equivalent to random erasure, which significantly improves both communication efficiency and the robustness of FL. Theoretical analysis and experimental results consistently demonstrate the effectiveness of our designs.

The out-of-distribution (OOD) generalization problem in federated learning (FL) has recently attracted significant research interest. A common approach, derived from centralized learning, is to extract causal features which exhibit causal relationships with the label. However, in FL, the global feature extractor typically captures only invariant causal features shared across clients and thus discards many other causal features that are potentially useful for OOD generalization. To address this problem, we propose FedUni, a simple yet effective architecture trained to extract all possible causal features from any input. FedUni consists of a comprehensive feature extractor, designed to identify a union of all causal feature types in the input, followed by a feature compressor, which discards potential \textit{inactive} causal features. With this architecture, FedUni can benefit from collaborative training in FL while avoiding the cost of model aggregation (i.e., extracting only invariant features). In addition, to further enhance the feature extractor's ability to capture causal features, FedUni add a causal intervention module on the client side, which employs a counterfactual generator to generate counterfactual examples that simulate distributions shifts. Extensive experiments and theoretical analysis demonstrate that our method significantly improves OOD generalization performance.

Dense video captioning aims to generate descriptive sentences for each temporally localized event in a video. This task comprises two subtasks: event detection and event captioning. Existing methods commonly adopt a DETR-like (Detection Transformer) architecture to perform both subtasks in parallel. These methods assume that both subtasks require the same visual information and thus extract a single event representation for each event using a shared query. We observe that event detection and event captioning emphasize different regions of a video. In particular, compared to event captioning, event detection tends to focus more on the boundary regions of event proposals. Therefore, relying on shared queries may hinder the ability of the model to meet the specific needs of each subtask, leading to suboptimal performance. In this paper, we propose decoupling the two subtasks by assigning distinct queries to each, enabling more accurate capture of task-specific features. Specifically, we introduce a task-specific query transformation module. This module utilizes two sets of task-specific prompts to transform shared queries into queries tailored for each subtask. These task-specific queries enable each subtask to attend to the video regions that are most beneficial to its respective objectives. By integrating our method into several state-of-the-art frameworks, we achieve superior performance on both event detection and event captioning.

Federated Learning (FL) faces challenges due to data heterogeneity, which limits the global model’s performance across diverse client distributions. Personalized Federated Learning (PFL) addresses this by enabling each client to process an individual model adapted to its local distribution. Many existing methods assume that certain global model parameters are difficult to train effectively in a collaborative manner under heterogeneous data. Consequently, they localize or fine-tune these parameters to obtain personalized models. In this paper, we reveal that both the feature extractor and classifier of the global model are inherently strong, and the primary cause of its suboptimal performance is the mismatch between local features and the global classifier. Although existing methods alleviate this mismatch to some extent and improve performance, we find that they either (1) fail to fully resolve the mismatch while degrading the feature extractor, or (2) address the mismatch only post-training, allowing it to persist during training. This increases inter-client gradient divergence, hinders model aggregation, and ultimately leaves the feature extractor suboptimal for client data. To address this issue, we propose FedPFT, a novel framework that resolves the mismatch during training using personalized prompts. These prompts, along with local features, are processed by a shared self-attention-based transformation module, ensuring alignment with the global classifier. Additionally, this prompt-driven approach offers strong flexibility, enabling task-specific prompts to incorporate additional training objectives (\eg, contrastive learning) to further enhance the feature extractor. Extensive experiments show that FedPFT outperforms state-of-the-art methods by up to 5.07%, with further gains of up to 7.08% when collaborative contrastive learning is incorporated.

Mobile edge computing (MEC), along with device-to-device (D2D) assisted MEC (D-MEC), are promising technologies that could improve the quality-of-experience for mobile devices (MDs) by offloading their tasks to edge servers or nearby idle MDs. There is a popular trend to develop distributed task offloading algorithms using multi-agent reinforcement learning (MARL), whose adoption of central critics during training makes the offloading still size-sensitive. Therefore, this paper proposes a Size-Insensitive Task Offloading (SITOff) algorithm for D-MEC based on fully-distributed offloading without maintaining any central venue. Specifically, taking advantage of the inherent graph-like structure of D-MEC, SITOff adopts graphs to represent MDs’ states and relationships and form each MD's local knowledge about D-MEC through graph computation. Furthermore, considering the limitation of local knowledge in performing whole performance-oriented offloading, each MD utilizes D2D-transmitting to exchange knowledge with its neighbors and form a comprehensive knowledge about D-MEC to enhance the coordination of distributed offloading. Additionally, regarding the different impacts of neighbors’ knowledge, each MD leverages attention mechanisms to selectively learn its neighbors’ knowledge during knowledge-exchange. Extensive experimental results show the superiority of SITOff over state-of-the-art MARL-based offloading algorithms in D-MEC with various MDs, and the easy collaboration of SITOff with curriculum-learning for large-scale D-MEC offloading.

As a classic data processing tool, Principal Component Analysis (PCA) has been widely applied in various data analysis applications. To mitigate the high computational complexity of PCA on big data, distributed PCA methods have been extensively studied, which disperse the computational tasks across multiple computation units while guaranteeing the accuracy. For the scenarios of distributed PCA in wireless networks, as the data is originally dispersed across different locations, it is further required to reduce the communication cost of distributed PCA in networks, which however has been seldom studied. Reducing the communication cost of distributed PCA in wireless networks requires not only appropriately partitioning the computation of PCA, ensuring accuracy, but also effectively assigning the partitioned computations and routing strategies to the nodes. In this paper, we propose CD-PCA, a communication-efficient distributed PCA (CD-PCA) scheme. This scheme implements a transmission-benefit equipartition strategy for the network to facilitate high-accuracy distributed computation and designs novel routing strategies for nodes to execute the distributed PCA within each partitioned region. Extensive simulation results demonstrate that the proposed CD-PCA scheme can reduce transmission costs by over 30% on average compared to related methods and baseline approaches.

Personalized federated learning (PFL) allows clients to collaboratively train their personalized models to handle situations where data from different clients are not independent and identically distributed (non-IID). Previous PFL research implicitly assumes that clients benefit most from those with similar data distributions. Correspondingly, methods such as personalized weight aggregation assign higher weights to similar clients during aggregation. We pose a question: can a client benefit from other clients with dissimilar data distributions, and if so, how? This question is particularly relevant in scenarios with a high degree of non-IID, where clients have widely different distributions, and learning from only similar clients will result in a loss of knowledge from many other clients. We note that when dealing with clients with similar distributions, current methods tend to enforce their models to be close in the parameter space. It is reasonable to conjecture that a client can benefit from dissimilar clients if we allow their models to depart from each other. Based on this idea, we propose DiversiFed, which allows each client to learn from clients with diversified distribution. DiversiFed pushes personalized models of clients with dissimilar distributions apart in the parameter space while pulling together those with similar distributions. In addition, to achieve the above effect without using prior knowledge of distribution, we design a loss function that leverages model similarity to determine the degree of attraction and repulsion between any two models. Experiments on benchmark and medical datasets show that DiversiFed can outperform the state-of-the-art (SOTA) methods by up to 3.19%.

Personalized federated learning (PFL) is a popular distributed learning framework that allows clients to have different models and has many applications where clients' data are in different domains, including autonomous driving, traffic surveillance, and medical diagnosis. The typical model of a client in PFL features a global encoder trained by all clients to extract universal features from the raw data and personalized layers (e.g., a classifier) trained using the client's local data. Nonetheless, due to the differences between the data distributions of different clients (also known as, domain gaps), the universal features produced by the global encoder largely encompass numerous components irrelevant to a certain client's local task. Some recent PFL methods address the above problem by personalizing specific parameters within the encoder. However, these methods encounter substantial challenges attributed to the high dimensionality and nonlinearity of neural network parameter space. In contrast, the feature space exhibits a lower dimensionality, providing greater intuitiveness and interpretability as compared to the parameter space. To this end, we propose a novel PFL framework named FedPick. FedPick achieves PFL within the low-dimensional feature space by adaptively selecting task-relevant features for each client from the features generated by the global encoder based on its local data distribution. It presents a more accessible and interpretable implementation of PFL compared to those methods working in the parameter space. Extensive experimental results on multiple cross-domain datasets show that FedPick can effectively select task-relevant features for each client and improve model performance in cross-domain FL.

We propose 3DFaceSculptor, a general-purpose framework for interactive 3D face editing. Given a source 3D face mesh with semantic materials, and a user-specified semantic image, 3DFaceSculptor can accurately edit the source mesh following the shape guidance of the semantic image, while preserving the source topology as rigid as possible. Recent studies on generating 3D faces focus on learning neural networks to predict 3D shapes, which requires high-cost 3D training datasets. These learning-based methods are limited in compatibility and can only handle face styles involved in the training datasets. Unlike these methods, our 3DFaceSculptor is a non-training and common framework, which only requires supervision from readily-available semantic images, and is compatible with producing various face styles unlimited by datasets. In 3DFaceSculptor, based on the differentiable renderer technique, we deform the source face mesh according to the correspondences between semantic images and mesh materials. However, guiding complex 3D shapes with a simple 2D image incurs extra challenges, that is, the deformation accuracy, surface smoothness, geometric rigidity, and global synchronization of the edited mesh must be guaranteed. To address these challenges, we propose a hierarchical optimization architecture to balance the global and local shape features, and further propose various strategies and losses to improve properties of accuracy, smoothness, rigidity, and so on. Extensive experiments show that our 3DFaceSculptor is able to produce impressive results and has reached the state-of-the-art level.

Federated learning (FL) generally suffers significant communication overhead from high-traffic gradient synchronization. The majority of existing studies on this problem aim at compressing gradients under the premise of reliable transmission. While transmission reliability can be ensured via TCP by default, the notably increased latency and retransmitted packets are prohibitive for most clients in FL. To tackle this issue, we propose FedMDC, a retransmission-free compression framework for FL over packet lossy networks. Given clients’ limited resources, FedMDC adopts multiple description coding to encode gradients into redundant descriptions for erasure resilience simply through multiplying an overcomplete matrix; and then quantizes these descriptions for compression. To further reduce quantization distortion and computational overhead, a reduced decoding algorithm is developed by decoding the aggregation of all clients’ encodings in conjunction with a customized dither quantization design. Besides, FedMDC explicitly supports adaptive bitrates subject to clients’ heterogeneous communication budgets, which maximize resource utilization to facilitate distortion reduction and accelerate model convergence. Theoretical analysis and experimental results both demonstrate the effectiveness of our scheme.

Federated learning (FL) enables distributed training via periodically synchronizing model updates among participants. Communication overhead becomes a dominant constraint of FL since participating clients usually suffer from limited bandwidth. To tackle this issue, top-k based gradient compression techniques are broadly explored in FL context, manifesting powerful capabilities in reducing gradient volumes via picking significant entries. However, previous studies are primarily conducted on the raw gradients where massive spatial redundancies exist and positions of non-zero (top-k) entries vary greatly between gradients, which both impede the achievement of deeper compressions. Top-k may also degrade the performance of trained models due to biased gradient estimations. Targeting the above issues, we propose FedTC, a novel transform coding based compression framework. FedTC transforms gradients into a new domain with more compact energy distributions, which facilitates reducing spatial redundancies and biases in subsequent sparsification. Furthermore, non-zero entries across clients from different rounds become highly aligned in the transform domain, motivating us to partition the gradients into smaller entry blocks with various alignment levels to better exploit these alignments. Lastly, positions and values of non-zero entries are independently compressed in a block-wise manner with our customized designs, through which a higher compression ratio is achieved. Theoretical analysis and extensive experiments consistently demonstrate the effectiveness of our approach.

Deep learning models often suffer performance degradation when test data diverges from training data. Test-Time Adaptation (TTA) aims to adapt a trained model to the test data distribution using unlabeled test data streams. In many real-world applications, it is quite common for the trained model to be deployed across multiple devices simultaneously. Although each device can execute TTA independently, it fails to leverage information from the test data of other devices. To address this problem, we introduce Federated Learning (FL) to TTA to facilitate on-the-fly collaboration among devices during test time. The workflow involves clients (i.e., the devices) executing TTA locally, uploading their updated models to a central server for aggregation, and downloading the aggregated model for inference. However, implementing FL in TTA presents many challenges, especially in establishing inter-client collaboration in dynamic environment, where the test data distribution on different clients changes over time in different manners. To tackle these challenges, we propose a server-side Temporal-Spatial Aggregation (TSA) method. TSA utilizes a temporal-spatial attention module to capture intra-client temporal correlations and inter-client spatial correlations. To further improve robustness against temporal-spatial heterogeneity, we propose a heterogeneity-aware augmentation method and optimize the module using a self-supervised approach. More importantly, TSA can be implemented as a plug-in to TTA methods in distributed environments. Experiments on multiple datasets demonstrate that TSA outperforms existing methods and exhibits robustness across various levels of heterogeneity. The code is available at https://github.com/ZhangJiayuan-BUAA/FedTSA.

To address data heterogeneity, the key strategy of Personalized Federated Learning (PFL) is to decouple general knowledge (shared among clients) and client-specific knowledge, as the latter can have a negative impact on collaboration if not removed. Existing PFL methods primarily adopt a parameter partitioning approach, where the parameters of a model are designated as one of two types: parameters shared with other clients to extract general knowledge and parameters retained locally to learn client-specific knowledge. However, as these two types of parameters are put together like a jigsaw puzzle into a single model during the training process, each parameter may simultaneously absorb both general and client-specific knowledge, thus struggling to separate the two types of knowledge effectively. In this paper, we introduce FedDecomp, a simple but effective PFL paradigm that employs parameter additive decomposition to address this issue. Instead of assigning each parameter of a model as either a shared or personalized one, FedDecomp decomposes each parameter into the sum of two parameters: a shared one and a personalized one, thus achieving a more thorough decoupling of shared and personalized knowledge compared to the parameter partitioning method. In addition, as we find that retaining local knowledge of specific clients requires much lower model capacity compared with general knowledge across all clients, we let the matrix containing personalized parameters be low rank during the training process. Moreover, a new alternating training strategy is proposed to further improve the performance. Experimental results across multiple datasets and varying degrees of data heterogeneity demonstrate that FedDecomp outperforms state-of-the-art methods up to 4.9%. The code is available at https://github.com/XinghaoWu/FedDecomp

In personalized federated learning (PFL), it is widely recognized that achieving both high model generalization and effective personalization poses a significant challenge due to their conflicting nature. As a result, existing PFL methods can only manage a trade-off between these two objectives. This raises an interesting question: Is it feasible to develop a model capable of achieving both objectives simultaneously? Our paper presents an affirmative answer, and the key lies in the observation that deep models inherently exhibit hierarchical architectures, which produce representations with various levels of generalization and personalization at different stages. A straightforward approach stemming from this observation is to select multiple representations from these layers and combine them to concurrently achieve generalization and personalization. However, the number of candidate representations is commonly huge, which makes this method infeasible due to high computational costs. To address this problem, we propose DualFed, a new method that can directly yield dual representations correspond to generalization and personalization respectively, thereby simplifying the optimization task. Specifically, DualFed inserts a personalized projection network between the encoder and classifier. The pre-projection representations are able to capture generalized information shareable across clients, and the post-projection representations are effective to capture task-specific information on local clients. This design minimizes the mutual interference between generalization and personalization, thereby achieving a win-win situation. Extensive experiments show that DualFed can outperform other FL methods. Code is available at https://github.com/GuogangZhu/DualFed.

Federated learning is a distributed machine learning paradigm designed to protect user data privacy, which has been successfully implemented across various scenarios. In traditional federated learning, the entire parameter set of local models is updated and averaged in each training round. Although this full network update method maximizes knowledge acquisition and sharing for each model layer, it prevents the layers of the global model from cooperating effectively to complete the tasks of each client, a challenge we refer to as layer mismatch. This mismatch problem recurs after every parameter averaging, consequently slowing down model convergence and degrading overall performance. To address the layer mismatch issue, we introduce the FedPart method, which restricts model updates to either a single layer or a few layers during each communication round. Furthermore, to maintain the efficiency of knowledge acquisition and sharing, we develop several strategies to select trainable layers in each round, including sequential updating and multi-round cycle training. Through both theoretical analysis and experiments, our findings demonstrate that the FedPart method significantly surpasses conventional full network update strategies in terms of convergence speed and accuracy, while also reducing communication and computational overheads.

Manga is a fashionable Japanese-style comic form that is composed of black-and-white strokes and is generally displayed as raster images on digital devices. Typical mangas have simple textures, wide lines, and few color gradients, which are vectorizable natures to enjoy the merits of vector graphics, e.g., adaptive resolutions and small file sizes. In this paper, we propose MARVEL (MAnga’s Raster to VEctor Learning), a primitive-wise approach for vectorizing raster gray-level mangas by Deep Reinforcement Learning (DRL). Unlike previous learning-based methods which predict vector parameters for an entire image, MARVEL introduces a new perspective that regards an entire manga as a collection of basic primitives—stroke lines, and designs a DRL model to decompose the target image into a primitive sequence for achieving accurate vectorization. To improve vectorization accuracies and decrease file sizes, we further propose a stroke accuracy reward to predict accurate stroke lines, and a pruning mechanism to avoid generating erroneous and repeated strokes. Extensive subjective and objective experiments show that our MARVEL can generate impressive results and reaches the state-of-the-art level.

Cross-domain federated learning (FL), where data on local clients come from different domains, is a common case of FL. In such a cross-domain case, features extracted from the raw data of different clients deviate from each other in the feature space, leading to a so-called feature shift. This phenomenon can reduce feature discrimination and degrade the performance of the learned model. However, most existing FL methods are not specifically designed for the cross-domain setting. In this article, we propose a novel cross-domain FL method named AlignFed. In AlignFed, each client model consists of a personalized feature extractor and a shared lightweight classifier. The feature extractor maps the features to a consistent space by aligning them to identical global target points. Inspired by recent studies in contrastive learning, AlignFed regards points that are uniformly distributed on the hypersphere as global target points. It then pushes features toward global target points of their corresponding classes and away from those of other classes to improve feature discrimination. The shared classifier aggregates knowledge across clients over the consistent feature space, which can mitigate performance degradation caused by feature shift while reducing communication cost. We conduct convergence analysis and perform extensive experiments to evaluate AlignFed.

When doctors write a medical report, they first focus on key regions in the image, which usually contain abnormal information. Then, they write the report based on their experience and professional knowledge. Some existing medical report generation methods design corresponding schemes to model this process, but they still have some drawbacks. Some of them adopt the attention map to model the doctors’ image-reading behavior, which requires additional cropping and feature extraction operations, bringing many extra computations. Other methods manually construct knowledge graphs to model the doctors’ domain knowledge, which requires manual construction for different disease domains and has less generalization. Therefore, We propose DEKG, a doctors behavior aware and domain knowledge driven model, to improve the quality of generated reports by modeling the fixed pattern of doctors when writing reports. Specifically, feature clustering is utilized to model the doctors’ image-reading behavior that only brings fewer computations, and then we capture key regions with possible abnormalities. We also design a method to automatically construct domain knowledge graphs, which can quickly accomplish the construction process for different disease domains without human intervention. Extensive experiments on datasets from different disease domains demonstrate that DEKG achieves competitive results with state-of-the-art methods and can generate high-quality reports.

In the past decade, significant advancements have been made in utilizing deep learning for breast lesion segmentation. Recently, researchers have increasingly focused on harnessing the power of multiple modalities, recognizing its potential for enhancing segmentation performance. We observe that in clinical practice, many radiologists often rely on two types of ultrasound images, namely ultrasound (US) and contrast-enhanced ultrasound (CEUS) data for diagnosis. This motivates us to propose a multi-modal segmentation network, called as IMAN (Iterative Mutual-Aid Network), based on these two modalities. The architecture of IMAN adopts a novel hourglass shape, featuring two branches connected by an ‘X’ pathway. One branch is dedicated to processing CEUS data, while the other branch handles US data. Each branch generates segmentation results specific to its respective modality. The ’X’ pathway, realized by a margin mask generator module, serves as a bridge between these branches by forcing the segmentation results from one branch as additional input to the other. This head-to-tail pathway effectively facilitates mutual aid between the two modalities. In addition, we propose an iterative training policy during the training process to fully exploit the information from both US and CEUS data. Experimental results on a Breast-US-CEUS dataset comprising 169 samples demonstrate the effectiveness of IMAN, achieving Dice Similarity Coefficient of 83.96% and 81.16% for US images and CEUS videos, respectively. These scores surpass those obtained by many state-of-the-art segmentation methods. Furthermore, IMAN exhibits robust generalization capabilities across different segmentation structures.

Personalized federated learning (PFL) reduces the impact of non-independent and identically distributed (non-IID) data among clients by allowing each client to train a personalized model when collaborating with others. A key question in PFL is to decide which parameters of a client should be localized or shared with others. In current mainstream approaches, all layers that are sensitive to non-IID data (such as classifier layers) are generally personalized. The reasoning behind this approach is understandable, as localizing parameters that are easily influenced by non-IID data can prevent the potential negative effect of collaboration. However, we believe that this approach is too conservative for collaboration. For example, for a certain client, even if its parameters are easily influenced by non-IID data, it can still benefit by sharing these parameters with clients having similar data distribution. This observation emphasizes the importance of considering not only the sensitivity to non-IID data but also the similarity of data distribution when determining which parameters should be localized in PFL. This paper introduces a novel guideline for client collaboration in PFL. Unlike existing approaches that prohibit all collaboration of sensitive parameters, our guideline allows clients to share more parameters with others, leading to improved model performance. Additionally, we propose a new PFL method named FedCAC, which employs a quantitative metric to evaluate each parameter’s sensitivity to non-IID data and carefully selects collaborators based on this evaluation. Experimental results demonstrate that FedCAC enables clients to share more parameters with others, resulting in superior performance compared to state-of-the-art methods, particularly in scenarios where clients have diverse distributions. The code is integrated into our FL training framework: https://github.com/kxzxvbk/Fling.

Dense video captioning, with the objective of describing a sequence of events in a video, has received much attention recently. As events in a video are highly correlated, leveraging relationships among events helps generate coherent captions. To utilize relationships among events, existing methods mainly enrich event representations with their context, either in the form of vision (i.e., video segments) or combining vision and language (i.e., captions). However, these methods do not explicitly exploit the correspondence between these two modalities. Moreover, the video-level context spanning multiple events is not fully exploited. In this paper, we propose MRCap, a novel relationship-based model for dense video captioning. The key of MRCap is a multi-modal and multi-level event relationship module (MMERM). MMERM exploits the correspondence between vision and language at both the event level and the video level via contrastive learning. Experiments on ActivityNet Captions and YouCook2 datasets demonstrate that MRCap achieves state-of-the-art performance.

Federated learning (FL) trains a shared global model by periodically aggregating gradients from local devices. Communication overhead becomes a principal bottleneck in FL since participating devices usually suffer from limited bandwidth and unreliable connections in uplink transmission. To address this problem, the gradient compression methods based on compressed sensing (CS) theory have been put forward recently. However, most existing CS-based works compress gradients independently, ignoring the gradient correlations between participants or adjacent communication rounds, which constrains the achievement of higher compression rates. In view of the above observation, we propose a novel gradient compression scheme named FedDCS, guided by distributed compressed sensing (DCS) theory. Following the design philosophy of separate encoding and joint decoding in DCS, FedDCS compresses gradients for participants in each round separately while reconstructing them at the central server jointly via fully exploiting correlated gradients from the previous round, which are known as side information (SI). Benefiting from this design, reconstruction performance is significantly improved with fewer decoding errors also iterations under the identical compression rate, and the total uploading bits to achieve model convergence are considerably reduced. Theoretical analysis and extensive experiments conducted on MNIST and Fashion-MNIST both verify the effectiveness of our approach.

Federated learning (FL) is an emerging paradigm of distributed machine learning. However, when applied to wireless network scenarios, FL usually suffers from high communication cost because clients need to transmit their updated gradients to a server in every training round. Although many gradient compression techniques like sparsification and quantization are proposed, they compress clients’ gradients independently, without considering the correlations among gradients. In this paper, we propose SVDFed, a collaborative gradient compression framework for FL. SVDFed utilizes Singular Value Decomposition (SVD) to find a few basis vectors, whose linear combination can well represent clients’ gradients at a certain round. Due to the correlations among gradients, these basis vectors can still well approximate new gradients in many subsequent rounds. With the help of basis vectors, clients only need to upload the coefficients of the linear combination to the server, which greatly reduces communication cost. In addition, SVDFed leverages the classical PID (Proportional, Integral, Derivative) control to determine the proper time to update basis vectors to maintain their representation ability. Through experiments, we demonstrate that SVDFed outperforms existing gradient compression methods in FL. For example, compared to a popular gradient quantization method QSGD, SVDFed can reduce the communication overhead by 66 % and pending time by 99 %.

In many domain-specific monitoring applications of wireless sensor networks (WSNs), such as structural health monitoring (SHM), volcano tomography, and machine diagnosis, all the raw data in WSNs are required to be gathered to the sink where a specialized centralized algorithm is then executed to extract some global features or model parameters. To reduce the large-scale raw data transmission while guaranteeing the global feature quality, there are two kinds of solutions: one is in-network processing, which generally needs to distribute the centralized computation of feature extraction into networks. Another solution is compressive sensing (CS) followed with the feature extraction (called feature CS in this article). An interesting question is: for in-network processing and feature CS, which kind of solutions is more cost efficient to accomplish the task of feature extraction? This question is seldom studied. To answer it, we take the case of SHM with WSNs along with the classic feature extraction algorithm, i.e., the Eigen-system realization algorithm (ERA), and appropriately design two novel routes for in-network processing and feature CS, respectively. Both theoretical analysis of the two solutions’transmission cost and numerous simulations have been conducted. Based on the comparison results, we summarize some guidelines on the solution choice for different kinds of WSNs for SHM. In addition, we find that, instead of guaranteeing the quality of raw data reconstructed, CS with guaranteeing the quality of feature extracted is usually more meaningful and cost efficient.

Automatically generating radiology reports has recently made great progress, which reduces the workload of radiologists. The high similarity among radiology images in training datasets forces existing methods to focus more on common medical facts existing in many images, such as pleural effusion and heart size. However, distinctive medical facts in individual images, such as edema and soft tissue that are rare in the whole dataset, receive little attention or are even ignored. This result leads to reports generated by existing methods that lack descriptions of distinctive medical facts in individual images. In this paper, we propose TRGD, a Transformer with graph-based distinctive attention, to generate high-quality radiology reports automatically. In TRGD, we first design a graph network to extract the features representing common medical facts. Then we lower their proportion in the global semantic features representing complete medical facts to increase the influence of distinctive medical facts. Experimental results on two datasets IU X-Ray and MIMIC-CXR demonstrate that, by sufficiently focusing on distinctive medical facts in individual images, TRGD indeed generates high-quality radiology reports and achieves competitive results with state-of-the-art methods.

One vital challenge in federated learning (FL) is the statistical heterogeneity of data in different clients, which negatively affects the performance of the finally obtained model. One common approach to address this problem, called as personalized federated learning (PFL), is to train a personalized model for each client. A key design issue in PFL-based methods is determining which parts of the model should be personalized for each client. For example, one popular method in PFL is to personalize the batch normalization layers. In this paper, we propose ChannelFed, a new PFL-based method which personalizes the channel attention module. ChannelFed is designed based on the following observation: Channel attention assigns different weights to channels for different classes of data, which can be utilized to exploit knowledge of heterogeneous data from different clients. By keeping the channel attention module localized, ChannelFed enables clients to concentrate on client-specific channels. ChannelFed implements normalization across samples in the channel attention module to better fit for statistical heterogeneity scenarios. Experiments on CIFAR-10, Fashion-MNIST, and CIFAR-100 datasets demonstrate that ChannelFed outperforms other PFL methods under statistical heterogeneity scenarios.

Federated learning (FL) is an emerging machine learning paradigm where multiple distributed clients collaboratively train a model without centrally collecting their raw data. In FL setting, it is a common case that the data on local clients come from different domains, e.g., photos taken by different mobile phones can vary in intensity and contrast due to the difference of imaging parameters. In such a cross-domain case, features extracted from data of different clients deviate from each other in the feature space, leading to the so-called feature shift. The feature shift can reduce the discrimination of features and degrade the performance of the learned model. However, most existing FL methods are not particularly designed for cross-domain setting. In this paper, we propose a novel cross-domain FL method, named AlignFed. In AlignFed, the model on each client is separated to a personalized feature extractor and a shared classifier. The former extracts consistent features among clients by aligning features of different clients to some specific points in the feature space. The latter aggregates the knowledge across clients over the consistent feature space, which can mitigate the performance degradation caused by the feature shift in cross-domain FL. We conduct experiments on common-used multi-domain datasets, including Digits-Five, Office-Caltech10, and DomainNet. The experimental results demonstrate that AlignFed can outperform the state-of-art FL methods.

Data heterogeneity is one of the main challenges faced by federated learning (FL). Unlike traditional FL methods (e.g. FedAvg) which train a global model for all clients, personalized federated learning (PFL) can address the above problem by training a personalized model for each client. Current mainstream PFL researches first obtain a global model through collaborative training among all clients and then fine-tune the global model on each client's local data to obtain personalized models. However, this two-staged approach has a drawback: when the heterogeneity of different clients is large, the obtained final global model can deviate from the distributions of all clients, and therefore is not a good starting point for updating personalized models. In this paper, we propose pFedGF, a new PFL method based on gradient fusion. Different from traditional two-staged PFL, in each round of pFedGF, each client maintains two gradients simultaneously, a global gradient to capture information from all clients, and a local gradient that reflects the specific distribution of each client. The two gradients are fused to obtain the updated direction of the personalized model for each client. We carried out experiments on MNIST, FMNIST, and CIFAR-10 datasets. The results demonstrate that in the presence of data heterogeneity, pFedGF outperforms other PFL methods.

Due to the high maneuverability and flexibility, unmanned aerial vehicles (UAVs) have been considered as a promising paradigm to assist mobile edge computing (MEC) in many scenarios including disaster rescue and field operation. Most existing research focuses on the study of trajectory and computation-offloading scheduling for UAV-assisted MEC in stationary environments, and could face challenges in dynamic environments where the locations of UAVs and mobile devices (MDs) vary significantly. Some latest research attempts to develop scheduling policies for dynamic environments by means of reinforcement learning (RL). However, as these need to explore in high-dimensional state and action space, they may fail to cover in large-scale networks where multiple UAVs serve numerous MDs. To address this challenge, we leverage the idea of “divide-and-conquer” and propose HT3O, a scalable scheduling approach for large-scale UAV-assisted MEC. First, HT3O is built with neural networks via deep RL to obtain real-time scheduling policies for MEC in dynamic environments. More importantly, to make HT3O more scalable, we decompose the scheduling problem into two-layered subproblems and optimize them alternately via hierarchical RL. This not only substantially reduces the complexity of each subproblem, but also improves the convergence efficiency. Experimental results show that HT3O can achieve promising performance improvements over state-of-the-art approaches.

The visual storytelling (VIST) task aims at generating reasonable, human-like and coherent stories with the image streams as input. Although many deep learning models have achieved promising results, most of them do not directly leverage the sentiment information of stories. In this paper, we propose a sentiment-aware generative model for VIST called SentiStory. The key of SentiStory is a multi-layered sentiment extraction module (MLSEM). For a given image stream, the higher layer gives coarse-grained but accurate sentiments, while the lower layer of the MLSEM extracts fine-grained but usually unreliable ones. The two layers are combined strategically to generate coherent and rich visual sentiment concepts for the VIST task. Results from both automatic and human evaluations demonstrate that with the help of the MLSEM, SentiStory achieves improvement in generating more coherent and human-like stories.

As sharing of accounts is quite common among family members or roommates, the design of efficient recommender schemes for shared accounts has raised much attention recently. Generally speaking, after each login, it is essential for a recommender system to identify the current user behind and leverage this information to make recommendations. One naive approach is first to identify the identity of the current user and then make recommendations. However, this two-stage based approach may not achieve satisfactory performance. The key is that the recommended items favoring identifying users in the first stage may not be interesting to the users, which can deplete the user's patience quickly and cause early termination of users. To address the problem, we propose a novel recommendation scheme that makes a tradeoff between recommending discriminating items (helpful for identifying the user) and recommending interesting ones to the user (helpful for increasing the number of clicks). Under this scheme, we develop a patience model to capture the user's dynamic patience level during the recommendation process. Moreover, considering the increasing popularity of mobile devices, we also incorporate mobile sensor data (i.e., angle, accelerometer, gyroscope, etc.) into our approach to further improve the performance of the system. We implemented the above system in an App on mobile devices and carried out extensive experiments. The results demonstrate that our proposed scheme significantly outperforms the existing state-of-the-art approaches.

Manga is a world popular comic form originated in Japan, which typically employs black-and-white stroke lines and geometric exaggeration to describe humans' appearances, poses, and actions. In this paper, we propose MangaGAN, the first method based on Generative Adversarial Network (GAN) for unpaired photo-to-manga translation. Inspired by the drawing process of experienced manga artists, MangaGAN generates geometric features and converts each facial region into the manga domain with a tailored multi-GANs architecture. For training MangaGAN, we collect a new data-set from a popular manga work with extensive features. To produce high-quality manga faces, we propose a structural smoothing loss to smooth stroke-lines and avoid noisy pixels, and a similarity preserving module to improve the similarity between domains of photo and manga. Extensive experiments show that MangaGAN can produce high-quality manga faces preserving both the facial similarity and manga style, and outperforms other reference methods.

The performance of deep learning models generally relies on large and high-quality labeled datasets. However, in medical domain, as labeling process is much more laborious and time-consuming, most medical datasets are much smaller compared with natural image datasets. To mitigate this weakness, recent researches in medical image analysis adopt semi-supervised learning methods, especially consistency regularization methods to learn from a large amount of unlabeled medical data. However, as these semi-supervised learning methods are originally designed for tasks of natural images, specific properties of medical domain are not fully investigated and utilized. In this paper, we present DK-Consistency, a domain knowledge guided consistency regularization method for semi-supervised breast cancer diagnosis in ultrasound images. In DK-Consistency, domain knowledge of medical doctors is first incorporated into the generation process of perturbed samples for each unlabeled image. Then consistency regularization is adopted to force the model to make consistent predictions for unlabeled images and their perturbed samples. Extensive experiments demonstrate that, by injecting domain knowledge, DK-Consistency significantly improves the diagnostic performance of breast cancer and outperforms many state-of the-art semi-supervised methods.

Quick Response (QR) code is one of the most worldwide used two-dimensional codes. Traditional QR codes appear as random collections of black-and-white modules that lack visual semantics and aesthetic elements, which inspires the recent works to beautify the appearances of QR codes. However, these works adopt fixed generation algorithms and therefore can only generate QR codes with a pre-defined style. In this paper, combining the Neural Style Transfer technique, we propose a novel end-to-end method, named ArtCoder, to generate the stylized QR codes that are personalized, diverse, attractive, and scanning-robust. To guarantee that the generated stylized QR codes are still scanning-robust, we propose a Sampling-Simulation layer, a module-based code loss, and a competition mechanism. The experimental results show that our stylized QR codes have high-quality in both the visual effect and the scanning-robustness, and they are able to support the real-world application.

The past year has witnessed the rapid development of applying the Transformer module to vision problems. While some researchers have demonstrated that Transformer-based models enjoy a favorable ability of fitting data, there are still growing number of evidences showing that these models suffer over-fitting especially when the training data is limited. This paper offers an empirical study by performing step-by-step operations to gradually transit a Transformer-based model to a convolution-based model. The results we obtain during the transition process deliver useful messages for improving visual recognition. Based on these observations, we propose a new architecture named Visformer, which is abbreviated from the ‘Vision-friendly Transformer’. With the same computational complexity, Visformer outperforms both the Transformer-based and convolution-based models in terms of ImageNet classification accuracy, and the advantage becomes more significant when the model complexity is lower or the training set is smaller. The code is available at https://github.com/danczs/Visformer.

Quick Response (QR) code is a popular form of matrix barcodes that are widely used to tag online links on print media (e.g., posters, leaflets, and books). However, standard QR codes typically appear as noise-like black/white squares (named modules) which seriously disrupt the attractiveness of their carriers. In this paper, we propose StyleCode-Net, a method to generate novel art-style QR codes which can better match the entire style of their carriers to improve the visual quality. For endowing QR codes with artistic elements, a big challenge is that the scanning-robustness must be preserved after transforming colors and textures. To address these issues, we propose a module-based deformable convolutional mechanism (MDCM) and a dynamic target mechanism (DTM) in StyleCode-Net. MDCM can extract the features of black and white modules of QR codes respectively. Then, the extracted features are fed to DTM to balance the scanning-robustness and the style representation. Extensive subjective and objective experiments show that our art-style QR codes have reached the state-of-the-art level in both visual quality and scanning-robustness, and these codes have the potential to replace standard QR codes in real-world applications.

Software-Defined-Networking (SDN) is progressively dominating the dynamic management for timely network trouble shooting and fine grained traffic scheduling in data center networks. One critical issue in SDN is to reduce the communication overhead between the switches and the controller. Such overhead is mainly caused by handling miss-match packets, because for each miss-match packet, a switch will send a request to the controller asking for forwarding rule. Existing approaches to address this problem generally need to deploy intermediate proxy or authority switches to hold rule copies, so as to reduce the number of requests sent to the controller. In this paper, we argue that using the intrinsic buffer in a SDN switch can also greatly reduce the communication overhead without using additional devices. If a switch buffers each miss-match packet, only a few header fields instead of the entire packet are required to be sent to the controller. Experiment results show that this can reduce 78.7 percent control traffic and 37 percent controller overhead at the cost of increasing only 5.6 percent switch overhead on average. If the proposed flow-granularity buffer mechanism is adopted, only one request message needs to be sent to the controller for a new flow with many arrival packets. Thus the control traffic and controller overhead can be further reduced by 64 percent and 35.7 percent respectively on average without increasing the switch overhead.

Large-scale canals with cascaded pools are constructed wordwide to divert water from rich to arid areas to mitigate water shortages. Efficient control of canals is essential to improve water-diversion performance. Numerous model-based approaches have been proposed and made great progress for canal control. However, when the predictive model is unavailable or unpromising for long time step predictions, model-free approaches could be considered as a possible way to achieve efficient control. Since most existing model-free approaches are focused on control of small canals or reservoirs, this article proposes a new control approach named policy and action reinforcement learning (PARL) for large-scale canals. We leverage the idea of “divide and conquer” to decompose the control task of large-scale canals into policy learning and action learning subtasks, and develop PARL by means of hierarchical reinforcement learning. Extensive experiments are conducted via numerical simulation on the case study of Chinese South to North Water Transfer Project, and experimental results show that PARL can achieve desirable performance improvements over other model-free learning approaches.

Online banks may disclose consumers’ shopping preferences due to various attacks. With differential privacy, each consumer can disturb his consumption amount locally before sending it to online banks. However, directly applying differential privacy in online banks will incur problems in reality because existing differential privacy schemes do not consider handling the noise boundary problem. In this paper, we propose an Optimized Differential prIvate Online tRansaction scheme (O-DIOR) for online banks to set boundaries of consumption amounts with added noises. We then revise O-DIOR to design a RO-DIOR scheme to select different boundaries while satisfying the differential privacy definition. Moreover, we provide in-depth theoretical analysis to prove that our schemes are capable to satisfy the differential privacy constraint. Finally, to evaluate the effectiveness, we have implemented our schemes in mobile payment experiments. Experimental results illustrate that the relevance between the consumption amount and online bank amount is reduced significantly, and the privacy losses are less than 0.5 in terms of mutual information.

Group detection is gaining popularity as it enables variousXzX applications ranging from marketing to urban planning. Existing methods use received signal strength indicator (RSSI) to detect co-located people as groups. However, this approach might have difficulties in crowded urban spaces since many strangers with similar mobility patterns could be identified as groups. Moreover, RSSI is vulnerable to many factors like the human body attenuation and thus is unreliable in crowded scenarios. In this work, we propose a behavior-aware group detection system (BaG). BaG fuses people’s mobility information and smartphone usage behaviors. We observe that people in a group tend to have similar phone usage patterns. Those patterns could be effectively captured by the proposed feature: number of bursts (NoB). Unlike RSSI, NoB is more resilient to environmental changes as it only cares about receiving packets or not. Besides, both mobility and usage patterns correspond to the same underlying grouping information. We propose a detection method based on collective matrix factorization to reveal the hidden associations by factorizing mobility information and usage patterns simultaneously. Experimental results indicate BaG outperforms baseline approaches by$3.97\% \sim 15.79\%$in F-score. The proposed system could also achieve robust and reliable performance in scenarios with different levels of crowdedness.

In recent years, deep learning has been widely used in breast cancer diagnosis, and many high-performance models have emerged. However, most of the existing deep learning models are mainly based on static breast ultrasound (US) images. In actual diagnostic process, contrast-enhanced ultrasound (CEUS) is a commonly used technique by radiologists. Compared with static breast US images, CEUS videos can provide more detailed blood supply information of tumors, and therefore can help radiologists make a more accurate diagnosis. In this paper, we propose a novel diagnosis model based on CEUS videos. The backbone of the model is a 3D convolutional neural network. More specifically, we notice that radiologists generally follow two specific patterns when browsing CEUS videos. One pattern is that they focus on specific time slots, and the other is that they pay attention to the differences between the CEUS frames and the corresponding US images. To incorporate these two patterns into our deep learning model, we design a domain-knowledge-guided temporal attention module and a channel attention module. We validate our model on our Breast-CEUS dataset composed of 221 cases. The result shows that our model can achieve a sensitivity of 97.2% and an accuracy of 86.3%. In particular, the incorporation of domain knowledge leads to a 3.5% improvement in sensitivity and a 6.0% improvement in specificity. Finally, we also prove the validity of two domain knowledge modules in the 3D convolutional neural network (C3D) and the 3D ResNet (R3D).

Automatic designing computationally efficient neural networks has received much attention in recent years. Existing approaches either utilize network pruning or leverage the network architecture search methods. This paper presents a new framework named network adjustment, which considers network accuracy as a function of FLOPs, so that under each network configuration, one can estimate the FLOPs utilization ratio (FUR) for each layer and use it to determine whether to increase or decrease the number of channels on the layer. Note that FUR, like the gradient of a non-linear function, is accurate only in a small neighborhood of the current network. Hence, we design an iterative mechanism so that the initial network undergoes a number of steps, each of which has a small 'adjusting rate' to control the changes to the network. The computational overhead of the entire search process is reasonable, i.e., comparable to that of re-training the final model from scratch. Experiments on standard image classification datasets and a wide range of base networks demonstrate the effectiveness of our approach, which consistently outperforms the pruning counterpart. The code is available at https://github.com/danczs/NetworkAdjustment.

The existing deep CNN models on single image super-resolution processing are computationally-intensive in terms of memory usage and training time. In resources-limited platforms, it is desirable to consider developing light-weight models for super-resolution tasks. This paper proposes a parallel-group convolution, which uses 25% computation of the standard convolutions. With parallel-group convolutions, we develop an efficient light-weight convolutional neural network named MobileSR for super-resolution. Experimental results show that our proposed method achieves appreciable improvements over the state-of-the-art models with approximately 75% size reduction. The source code is available at https://github.com/DestinyK/MobileSR.

Online bank system has been widely deployed to provide financial service worldwide. Big data mining brings serious privacy issues to consumers because consumption records may disclose their activities. Existing authentication and encryption algorithms in privacy-preserving deposit transaction schemes are mostly limited to the case that each consumption has to be reported to online banks directly and an insider adversary can monitor a consumer's bank account during the financial transaction period. Motivated by the above problems, we propose a deposit transaction scheme based on the differential privacy. Considering a mobile payment system as a noise generator, this scheme aims to reduce the relevancy between real consumption and deposit transaction. In addition, we provide in-depth theoretical analysis that our scheme can satisfy the definition of the differential privacy. Experimental results illustrate the consumption pattern can be protected with the differential privacy.

Convolutional neural networks (CNNs) have achieved remarkable success in image recognition. Although the internal patterns of the input images are effectively learned by the CNNs, these patterns only constitute a small proportion of useful patterns contained in the input images. This can be attributed to the fact that the CNNs will stop learning if the learned patterns are enough to make a correct classification. Network regularization methods like dropout and SpatialDropout can ease this problem. During training, they randomly drop the features. These dropout methods, in essence, change the patterns learned by the networks, and in turn, forces the networks to learn other patterns to make the correct classification. However, the above methods have an important drawback. Randomly dropping features is generally inefficient and can introduce unnecessary noise. To tackle this problem, we propose SelectScale. Instead of randomly dropping units, SelectScale selects the important features in networks and adjusts them during training. Using SelectScale, we improve the performance of CNNs on CIFAR and ImageNet.

Pyroelectric infrared (PIR) sensors are considered to be promising devices for device-free localization due to its advantages of low cost, energy efficiency, and the immunity from multi-path fading. However, most of the existing PIR-based localization systems only utilize the binary information of PIR sensors and therefore require a large number of carefully deployed PIR sensors. A few works directly map the raw data of PIR sensors to one's location using machine learning approaches. However, these data-driven approaches require abundant training data and suffer from environmental change. In this paper, we propose PIRATES, a PIR-based device-free localization system based on the raw data of PIR sensors. The key of PIRATES is to extract a new type of location information called azimuth change. The extraction of the azimuth change relies on the physical properties of PIR sensors. Therefore, no abundant training data are needed and the system is robust to environmental change. Through experiments, we demonstrate that PIRATES can achieve higher localization accuracy than the state-of-the-art approaches. In addition, the information of the azimuth change can be easily incorporated with other information of PIR signals (e.g. amplitude) to improve the localization accuracy.

Screen lock is a critical security feature for smart-phones to prevent unauthorized access. Although various screen unlocking technologies including fingerprint and facial recognition have been widely adopted, they still have some limitations. For example, fingerprints can be stolen by special material stickers and facial recognition systems can be cheated by 3D-printed head models. In this paper, we propose EmgAuth, a novel electromyography(EMG)-based smartphone unlocking system based on the Siamese network. EmgAuth leverages the Myo armband to collect the EMG data of smartphone users and enables users to unlock their smartphones when picking up and watching their smartphones. In particular, when training the Siamese network, we design a special data augmentation technique to make the system resilient to the rotation of the armband. We conduct experiments including 40 participants and the evaluation results show that EmgAuth can effectively authenticate users with an average true acceptance rate of 91.81% while keeping the average false acceptance rate of 7.43%. In addition, we also demonstrate that EmgAuth can work well for smartphones with different sizes and at different locations, and is applicable for users with different postures. EmgAuth bears great promise to serve as a good supplement for existing screen unlocking systems to improve the safety of smartphones.

Wireless communication demands increase sharply with the explosive growth of mobile devices. The communications mainly depend on the infrastructure-based networks, e.g., WLANs and cellular networks. However, such wireless connections may be unavailable in crowded areas (e.g., concert and conference hall) or interrupted by infrastructure failures caused by earthquake or tsunami. These promote the evolution of local communication systems over device-to-device communication, such as Bluetooth and Wi-Fi Direct (WFD). However, none of the existing studies construct a full-featured local communication system, and they do not consider how to support the user mobility either. In this article, we implement and evaluate the performance of a WFD-based local communication system. First, we improve the intragroup communication by the native implementation of WFD on the Android platform, and propose an application-layer forwarding solution for the intergroup communication, which can be applied to three or more connected groups. Then, we put forward a self-adaptive handover mechanism taking user mobility and node failures into account. To deal with the uncertainty in the handover decision procedure, a fuzzy-logic-based normalized quantitative decision algorithm (FNQD) with the weights derived from the fuzzy analytic hierarchy process (FAHP) is utilized. Finally, we evaluate the performance of the system through both simulation and experiment analysis. Results show that we can get a maximum throughput of 31.7 Mb/s for the intragroup communication and a maximum goodput of 4.76 Mb/s for the intergroup communication. What is more, mobile devices could perform various types of handover according to their roles and status, which could improve the robustness of the local communication system.

Visitor counting can be attractive to various applications, like business management and marketing investigation. Recently, many studies have employed wireless signals to achieve visitor counting without people's active participation and privacy intrusion. However, existing systems mainly count the overall visitors inside a certain area, which fails to provide the fine-grained information of the coming and leaving visitor flow. Unlike previous studies, this article proposes to count the in-and-out visitors to monitor visiting frequency and population, which can be applied for many indoor places, such as shops and restaurants. Therefore, we present the first WiFi-based in-and-out visitor counting system, Door-Monitor, which obtains the direction (enter or exit) and the number of visitors passing by the door. The WiFi signals enable us to count the visitors in a low-cost and nonintrusive way, and it can tell the exact number of visitors even when multiple persons pass by the door simultaneously. To detect the visitors' passing direction, we show that the patterns in the phase difference series can indicate the entering and exiting passing directions by analyzing the effects of the passing behavior on the signal's phase information. To count the passing visitors, we perform a short time Fourier transformation on the phase difference series to generate the spectrogram, on which the convolutional neural network is applied for building a counting model. The experimental results show that the average accuracies of passing direction detection and visitor counting are 95.2% and 94.5%, respectively.

Human respiration state is an important indicator to reflect health conditions. Recent advances in wireless human sensing have enabled device-free respiration monitoring using narrow-band wireless signals, which, however, fail to map the estimated respiration states to multiple persons. In this paper, we present Multi-Breath, a UWB-based system to achieve separate respiration monitoring for multiple persons. The UWB radar can accurately measure the travelling distance of the signals, which helps to separate the signals affected by different persons and map the detected respiration patterns to the corresponding persons with the location information. However, the radar signal time series of each person are quite noisy due to the multi-path effects caused by the respiration movements of other persons, making it difficult to accurately estimate the respiration state. To overcome this challenge, we propose to transform the UWB radar signal matrices of different persons as separate RGB images to reveal the respiration pattern of each individual. Then, the image processing operations, including image smoothing, edge detection, dilation and erosion, are applied to identify the breathing cycles. Finally, the respiration state, including the respiration rate and the presence of apnea, is estimated via blob detection and calibration. Extensive experiments show that the mean absolute error on respiration rate estimation is 0.3 - 0.6 bpm, and the percentage of missed and false detected apnea is 3% - 7%.

Target-level sentiment classification aims to identify the sentiment of multiple targets in a sentence. Although existing approaches based on neural network have achieved good performance in this task, we find that many approaches tend to give the same predictions for instances that have multiple targets, and this tendency can make a low accuracy for those instances that have different classes for different targets. Based on this observation, we propose MemNetAR, a memory network which can explicitly leverage the adversative relation among multiple targets in a sentence. Specifically, we add an adversative loss to the cross-entropy loss when there are adversative words between targets. The experimental results on public laptop and restaurant datasets prove that our model can improve 0.84% and 0.8% on total test dataset, and improve 2.97% and 2.68% on the dataset consisting of those instances with multiple targets but different classes by leveraging this new adversative information.

Text clustering has been widely used in many Natural Language Processing (NLP) applications such as text summarization and news recommendation. However, most of the current algorithms need to predefine a clustering number, which is difficult to obtain. Moreover, the mutli-label clustering is useful in multiple clustering tasks in many applications, but related works are rarely available. Although several studies have attempted to solve above two problems, there is a need for methods that can solve the two issues simultaneously. Therefore, we propose a new text clustering algorithm called Word2Cluster. Word2Cluster can automatically generate an adaptive number of clusters and support multi-label clustering. To test the performance of Wrod2Cluster, we build a Chinese text dataset, Hotline, according to real world applications. To evaluate the clustering results better, we propose an improved evaluation method based on basic accuracy, precision and recall for multi-label text clustering. Experimental results on a Chinese text dataset (Hotline) and a public English text dataset (Reuters) demonstrate that our algorithm can achieve better F1-measure and runs faster than the state-of- the-art baselines.

Gesture recognition systems based on myoelectric signal have raised more and more attention from researchers. Traditional gesture recognition methods are susceptible to multiple types of noise and require a large number of features, which increase overhead and decrease recognition efficiency. Fully utilizing the characteristics of the signal to recognize gestures is a big challenge. This paper proposes an improved empirical mode decomposition method (XB-EMD) based on autocorrelation function to denoise myoelectric signal. In addition, a novel deep neural network (CRNet) which combines the CNN and RNN layers together is trained for classifying the gestures based on denoised myoelectric signal. Experimental results show that the proposed gesture recognition system can improve recognition effectiveness at 97.4% with 10 typical gestures, and 72.55% with 16 complicated gestures.

Recently, the automatic text summarization has been widely used in text compression tasks. The Attention mechanism is one of the most popular methods used in the seq2seq (Sequence to Sequence) text summarization models. The current attention mechanisms usually use the hidden states of the encoder and the decoder to generate attention distributions. However, they ignore the information of the word waiting to be input into the decoder, leading to possible failures to obtain accurate attention distributions. In this work, we propose a novel attention mechanism further adding the decoder inputs into the operation of generating attention distributions. To our best knowledge, this is the first time that the decoder input has been added to the process of calculating the attention vector. The attention mechanism we proposed to generate the attention distributions considers context similarities as well as semantic similarities, which is closer to the behavior of the human summarizer. We also applied our attention mechanism to the seq2seq based summarization model and trained it on a large corpus containing hundreds of thousands of article-summary pairs. The experimental results on two summarization datasets demonstrate that our attention mechanism outperforms the existing well-known ones. For the popular evaluation metric of the text summarization, our method obtains a 2.93 ROUGE-2 score relative gain compared with the popular attention mechanism Bahdanau Attention, and a 2.21 ROUGE-2 score improvement compared with the best baseline method Luong Attention.

Group detection is gaining popularity as it enables various applications ranging from marketing to urban planning. The group information is an important social context which could facilitate a more comprehensive behavior analysis. An example is for retailers to determine the right incentive for potential customers. Existing methods use received signal strength indicator (RSSI) to detect co-located people as groups. However, this approach might have difficulties in crowded urban spaces since many strangers with similar mobility patterns could be identified as groups. Moreover, RSSI is vulnerable to many factors like the human body attenuation and thus is unreliable in crowded scenarios. In this work, we propose a behavior-aware group detection system (BaG). BaG fuses people's mobility information and smartphone usage behaviors. We observe that people in a group tend to have similar phone usage patterns. Those patterns could be effectively captured by the proposed feature: number of bursts (NoB). Unlike RSSI, NoB is more resilient to environmental changes as it only cares about receiving packets or not. Besides, both mobility and usage patterns correspond to the same underlying grouping information. The latent associations between them cannot be fully utilized in conventional detection methods like graph clustering. We propose a detection method based on collective matrix factorization to reveal the hidden associations by factorizing mobility information and usage patterns simultaneously. Experimental results indicate BaG outperforms baseline approaches by in F-score. The proposed system could also achieve robust and reliable performance in scenarios with different levels of crowdedness.

With the blossoming of social networking platforms like Twitter and Facebook, how to infer the opinions of online social network users on specific topics they had not directly given yet, has received much attention. Existing solutions mainly rely on one's previous posted messages. However, recent studies show that over 40% of users opt to be silent all or most of the time and post very few messages. Consequently, the performance of existing solutions will drop dramatically when they are applied to infer silent users' opinions, and how to infer the opinions of these silent users becomes a meaningful while challenging task. Inspired by the collaborative filtering techniques in cold-start recommendations, we infer the opinions of silent users by leveraging the text content posted by active users and their relationships between silent users. Specifically, we first consider both observed and pseudo relationships among users, and cluster users into communities in order to extract various kinds of features for opinion inference. We then design a coupled sparse matrix factorization (CSMF) model to capture the complex relations among these features. Extensive experiments on real-world data from Twitter show that our CSMF model achieves over 80% accuracy for the inference of silent users' opinions.

People counting provides valuable information on population mobility and human dynamics, which plays a critical role for intelligent crowd control and retail management. Recently, people counting has been achieved via radio-frequency signals as human presence can influence the propagation of wireless signals, from which the information of the moving crowd can be extracted. However, most of the existing studies using wireless signals only apply to the scenario when people keep moving all the time. Besides, they require labour-intensive training phase for building the counting model. In the Wi-Count system, we take another approach, which is to count the people passing by the doorway with COTS WiFi devices. It can not only detect the passing direction, but also identify the number of people even when multiple persons pass by concurrently without regulating passing behavior and pre-trained counting model. The passing direction is recognized by modeling the effects of the bi-directional passing behavior on the phase difference of WiFi signals. In addition, the number of passing people is obtained through an enhanced signal separation algorithm for providing precise counting result. Extensive experiments show the average accuracy on passing direction detection and passing people counting are about 95% and 92% respectively.

Recently, non-intrusive respiration monitoring has attracted much attention. Many respiration monitoring systems using the commercial off-the-shelf WiFi devices have been developed. However, these systems mainly have difficulties in the presence of multiple persons. The difficulty generally comes from the separation of the effects of multiple persons' respiration on the received WiFi signals. Another problem is that even though the separation can be feasible with some complicated algorithms, it is still impossible to map the multiple identified respiration states to the corresponding persons. In this paper, we study the problem of multi-person sleeping respiration monitoring and try to address the above challenges. Instead of focusing on developing complicated signal processing algorithms, we take another approach: via the deployment of WiFi transceivers. The key insight comes from the WiFi Fresnel zone model, which indicates that a carefully placed WiFi transceiver may only be affected by the person in a certain location. Furthermore, we consider the sleeping movements of people as well as the sleeping posture change to improve the robustness of the system. Extensive experiments show that we can successfully estimate the respiration rate of multiple persons, with the Mean Absolute Error (MAE) of 0.5 bpm - 1 bpm.

Detecting shopping groups is gaining popularity as it enables various applications ranging from marketing to advertising. Existing methods exploit WiFi probe requests to detect shopping groups by identifying co-located customers. However, the probe request is prone to suffer from device heterogeneity which might pose a severe data sparseness problem. More importantly, we find that a certain amount of shopping groups would separate sometimes which makes traditional methods unreliable. In this paper, we propose a shopping group detection system using WiFi (SNOW). Instead of collecting probe requests, SNOW utilizes the WiFi data from smartphones associated with the deployed access points (APs). We could thus obtain data from different devices and even ensure a data granularity of seconds using Arping. Besides, we exploit an effective heuristic extracted from two observations of shopping group dynamics to improve the detection performance. First, the probability of group separation differs in diverse areas. Second, the proportion of group participation and individual engagement differs in different activities of the mall. Therefore, APs under which shopping groups appear more frequently and barely separate should contribute more in measuring customer similarity. Lastly, we represent the measured similarity into a matrix format and apply matrix factorization with a sparsity constraint to derive grouping results directly. According to our experiments in a large shopping mall, SNOW improves the detection performance of baseline approaches by 13.2% on average.

One critical issue in SDN is to reduce the communication overhead between the switches and the controller. Such overhead is mainly caused by handling miss-match packets, because for each miss-match packet, a switch will send a request to the controller asking for forwarding rule. Existing approaches to address this problem generally need to deploy intermediate proxy or authority switches to hold rule copies, so as to reduce the number of requests sent to the controller. In this paper, we argue that using the intrinsic buffer in a SDN switch can also greatly reduce the communication overhead without using additional devices. If a switch buffers each miss-match packet, only a few header fields instead of the entire packet are required to be sent to the controller. Experiment results show that this can reduce 78.7% control traffic and 37% controller overhead at the cost of increasing only 5.6% switch overhead on average. If the proposed flow-granularity buffer mechanism is adopted, only one request message needs to be sent to the controller for a new flow with many arrival packets. Thus the control traffic and controller overhead can be further reduced by 64% and 35.7% respectively on average without increasing the switch overhead.

The security issue of public WiFi is gaining more and more concern. By listening to probe requests, an adversary can obtain the SSID list of the APs to which a smartphone previously connected, and utilizes this information to trick the smartphone into associating to it. However, with the enhancement of security level, most smartphones now do not proactively disclose their SSID lists, making these attacks obsolete. In this paper, we propose City-Hunter, an attacker that can lure nearby smartphones without knowing their SSID information. City-Hunter establishes and maintains an SSID database by integrating both offline and online information. Meanwhile, it smartly chooses some SSIDs to hit a smartphone according to the past record and freshness. We evaluate the performance of City-Hunter in different public places. The results demonstrate that City-Hunter is able to successfully hit 12% ~ 18% smartphones without knowing their SSID information, which is about 4 ~ 8 times improvement compared to the similar attacks like KARMA and MANA.

The performance of GPUs on mobile devices is generally the bottleneck of multimedia mobile applications (e.g., 3D games and virtual reality). Previous attempts to tackle the issue mainly migrate GPU computation to servers residing in remote cloud centers. However, the costly network delay is especially undesirable for highly-interactive multimedia applications since a fast response time is critical for user experience. In this paper, we propose GBooster, a system that accelerates multimedia mobile applications by transparently offloading GPU tasks onto neighboring multimedia devices such as Smart TVs and Gaming Consoles. Specifically, GBooster intercepts and redirects system graphics calls by utilizing the Dynamic Linker Hooking technique, which requires no modification of the applications and the mobile systems. In addition, a major concern for offloading is the high energy consumption incurred by network transmissions. To address this concern, GBooster is designed to intelligently switch between the low-power Bluetooth and the high-throughput WiFi based on the traffic demand. We implement GBooster on the Android system and evaluate its performance. The results demonstrate that it can boost applications' frame rates by up to 85%. In terms of power consumption, GBooster can preserve up to 70% energy compared with local execution.

Recently with the sharing economy of WiFi, IEEE 802.11 Wireless Mesh Networks (WMNs) have been deployed exponentially in more places. With the booming increase of density of APs as well as mobile clients (MCs), the handoff between MCs and APs becomes more frequent than ever before. One problem of most existing network-assisted fast handoff mechanisms is that an AP needs to switch to different channel to probe MCs, thus disrupting its current communications. In this paper, we propose a fast handoff mechanism called OppoScan. OppoScan opportunistically leverages nearby MCs and APs to produce the required information of neighboring AP for handoff, thus significantly decrease the number of switching channel of APs. We have implemented the OppoScan in about 200 APs in two shopping malls and the results demonstrated the performance of the proposed scheme.

Nowadays, with the help of various on-line social networking applications, one can freely interact with any person in their friends list. However, in most conditions, people in your friends list are those you know a priori, either via face-to-face talking or introducing by a common friend. This way of establishing ones friends list can miss many potential friends nearby. These `potential real life friends nearby are those who are invisible to you (you do not know before), physically close to you (e.g. taking the same subway for commuting, doing exercises at the same time slot), and share the same interest with you. Correspondingly, we developed Wi-friend. Wi-friend, once installed in your mobile phone, can help to find these `potential friends nearby'. Using Wi-friend, one only needs to briefly specify his/her interest. Then the smartphone starts to look for nearby people with the same interest and add each other into ones friends list automatically. One special feature of Wi-friend is that it does not rely on Internet connection. We leverage the Wi-Fi tethering technique, and let ones smartphone to switch between Wi-Fi hot-spot mode and Wi-Fi client mode to exchange information with nearby people. In addition, Wi-friend can dynamically determine how long to stay in each mode to maximize the number of people who can exchange information. Extensive experiments and simulations demonstrated the technical feasibility and the effectiveness of our Wi-friend system.

As an alternative to current wired-based networks, wireless sensor networks (WSNs) are becoming an increasingly compelling platform for engineering structural health monitoring (SHM) due to relatively low-cost, easy installation, and so forth. However, there is still an unaddressed challenge: the application-specific dependability in terms of sensor fault detection and tolerance. The dependability is also affected by a reduction on the quality of monitoring when mitigating WSN constrains (e.g., limited energy, narrow bandwidth). We address these by designing a dependable distributed WSN framework for SHM (called DependSHM) and then examining its ability to cope with sensor faults and constraints. We find evidence that faulty sensors can corrupt results of a health event (e.g., damage) in a structural system without being detected. More specifically, we bring attention to an undiscovered yet interesting fact, i.e., the real measured signals introduced by one or more faulty sensors may cause an undamaged location to be identified as damaged (false positive) or a damaged location as undamaged (false negative) diagnosis. This can be caused by faults in sensor bonding, precision degradation, amplification gain, bias, drift, noise, and so forth. In DependSHM, we present a distributed automated algorithm to detect such types of faults, and we offer an online signal reconstruction algorithm to recover from the wrong diagnosis. Through comprehensive simulations and a WSN prototype system implementation, we evaluate the effectiveness of DependSHM.

Internet of things (IOT) is emerging as sensing paradigms in many domain-specific monitoring applications in smart cities, such as structural health monitoring (SHM) and smart grid monitoring. Due to the large size of the monitoring objects (e.g., civil structure or the power grid), plenty of sensors need to be deployed and organized to be a large scale of multihop wireless sensor networks (WSNs), which tends to have quite high transmission cost. In-network processing is an efficient way to reduce the transmission cost in WSNs. However, implementing in-network processing for above domain-specific monitoring usually requires to losslessly distribute a dedicate domain-specific algorithm into WSNs, which is much different from most existing in-network processing works. This paper conducts a case study of a classic centralized SHM algorithm, i.e., eigensystem realization algorithm (ERA), and shows how to losslessly and optimally in-network process ERA, especially the typical feature extraction method, i.e., that is singular value decomposition (SVD) therein, in a WSN. Based on whether the intermediate data can be processed together or not by sensor nodes, we respectively implement tree-based in-network processing of SVD and chain-based in-network processing of SVD in WSNs. We prove that using an appropriate shallow light tree as routes for tree-based in-network processing of SVD, can achieve the approximation ratio ${\text{1}}+\sqrt{2}$ (in terms of transmission cost), while for the chain-based in-network processing of SVD, we design two efficient heuristic algorithms for searching the optimal routes. Extensive simulation results validate the efficiency of these proposed schemes that are customized for SVD-based IOT applications.

Wireless networks offer many advantages over wired local area networks such as scalability and mobility. Strategically deployed wireless networks can achieve multiple objectives like traffic offloading, network coverage, and indoor localization. To this end, various mathematical models and optimization algorithms have been proposed to find optimal deployments of access points (APs). However, wireless signals can be blocked by the human body, especially in crowded urban spaces. As a result, the real coverage of an on-site AP deployment may shrink to some degree and lead to unexpected dead spots (areas without wireless coverage). Dead spots are undesirable, since they degrade the user experience in network service continuity, on one hand, and, on the other hand paralyze some applications and services like tracking and monitoring when users are in these areas. Nevertheless, it is nontrivial for existing methods to analyze the impact of human beings on wireless coverage. Site surveys are too time consuming and labor intensive to conduct. It is also infeasible for simulation methods to predict the number of on-site people. In this article, we propose DMAD, a Data-driven Measuring of Wi-Fi Access point Deployment, which not only estimates potential dead spots of an on-site AP deployment but also quantifies their severity, using simple Wi-Fi data collected from the on-site deployment and shop profiles from the Internet. DMAD first classifies static devices and mobile devices with a decision-tree classifier. Then it locates mobile devices to grid-level locations based on shop popularities, wireless signal, and visit duration. Last, DMAD estimates the probability of dead spots for each grid during different time slots and derives their severity considering the probability and the number of potential users. The analysis of Wi-Fi data from static devices indicates that the Pearson Correlation Coefficient of wireless coverage status and the number of on-site people is over 0.7, which confirms that human beings may have a significant impact on wireless coverage. We also conduct extensive experiments in a large shopping mall in Shenzhen. The evaluation results demonstrate that DMAD can find around 70% of dead spots with a precision of over 70%.

Wireless charging technology is considered as a promising solution to address the energy limitation problem for wireless sensor networks (WSNs). In scenarios where the deployed chargers are static, we generally require a number of chargers to work simultaneously. However, due to the radio interference among different wireless chargers, scheduling these chargers is generally necessary. This scheduling problem is challenging since each charger's charging utility cannot be calculated independently due to the nonlinear superposition charging effect caused by radio interference. In this paper, based on the concurrent charging model, we formulate the concurrent charging scheduling problem (CCSP) with the objective of quickly fully charging all the sensor nodes. After proving the NP-hardness of CCSP, we propose two efficient greedy algorithms, and give the approximation ratio of one of them. Both the two greedy algorithms' performances are very close to that of a well-designed genetic algorithm (GA) which performs almost as well as a brute force algorithm at small network and charger scale. However, the running time of the two greedy algorithms is far lower than that of the GA. We conduct extensive simulations and specially implemented a testbed for wireless chargers. The results verified the good performance of the proposed algorithms.

Texting-while-driving (T&D) is one of the top dangerous behaviors for drivers. Many interesting systems and mobile phone applications have been designed to help to detect or combat T&D. However, for a T&D detection system to be practical, a key property is its capability to distinguish driver's mobile phone from passengers'. Existing solutions to this problem generally rely on the user's manual input, or utilize specific localization devices to determine whether a mobile phone is at the driver's location. In this paper, we propose a method which is able to detect T&D automatically without using any extra devices. The idea is very simple: when a user is composing messages, the smartphone embedded sensors (i.e., gyroscopes, accelerometers, and GPS) collect the associated information such as touchstrokes, holding orientation and vehicle speed. This information will then be analyzed to see whether there exists some specific T&D patterns. Extensive experiments have been conducted by different persons and in different driving scenarios. The results show that our approach can achieve a good detection accuracy with low false positive rate. Besides being infrastructurefree and with high accuracy, the method does not access the content of messages and therefore is privacy-preserving.

It is well recognized that walking while using mobile phones will make people more susceptible at various risks. Existing works to improve smartphone users' safety are mainly limited to detecting incoming vehicles. They are not able to address some more common and equally dangerous accidents such as trips, falling from stairs, platforms, or falling into an open manhole. These hazards are generally caused by a sudden change of ground. In this paper, we propose InfraSee, a system that is able to detect sudden change of ground for pedestrian mobile phone users. InfraSee augments smartphones with a small infrared sensor which measures the distance of the ground surface from the sensor. The temporal variation of distance can provide information about the change of ground surface ahead. InfraSee also leverages the information of smartphone sensors to improve detection accuracy, to reduce energy consumption, and to avoid unnecessary alarms. We have carried out extensive experiments in different scenarios and by different users. The results show that InfraSee is able to reliably detect about 80 percent change of ground surfaces. In addition, InfraSee can reliably identify the awareness of smartphone users and reduce unnecessary alarms.

In many domain-specific monitoring applications of wireless sensor networks (WSNs), such as structural health monitoring, volcano tomography, and machine diagnosis, the raw data in WSNs are required to be losslessly gathered to the sink, where a specialized centralized algorithm is then executed to extract some global features or model parameters. To reduce the large raw data transmission, in-network processing is usually employed. However, different from most existing in-network processing works that pre-assume some common computation/aggregation functions, in-network processing of a given centralized algorithm requires exact partitioning of the algorithm first and then appropriately assigning the partitioned computations into WSNs. We call this lossless in-network processing, which has not been studied much. Lossless in-network processing raises two questions: 1) what pattern should a centralized algorithm be partitioned into so that the partitioned computations can be flexibly assigned into a WSN with arbitrary topology? and 2) for each partition pattern, how should efficient routing for the resource-limited sensor nodes be designed? These two questions can be referred to as a topology-constrained computation partition problem and a computation-constrained routing design problem, respectively. In this paper, we first introduce some general patterns on the topology-constrained computation partition. Then, with the computation constraints in the patterns, we present a series of novel routing schemes customized for different cases of computation results. The work in this paper can also serve as a guideline for distributed computing of big data, where the data spreads in a large network.

Smart devices are increasingly shrinking in size, which results in new challenges for user-mobile interaction through minuscule touchscreens. Existing works to explore alternative interaction technologies mainly rely on external devices which degrade portability. In this paper, we propose UbiTouch, a novel system that extends smartphones with virtual touchpads on desktops using built-in smartphone sensors. It senses a user's finger movement with a proximity and ambient light sensor whose raw sensory data from underlying hardware are strongly dependent on the finger's locations. UbiTouch maps the raw data into the finger's positions by utilizing Curvilinear Component Analysis and improve tracking accuracy via a particle filter. We have evaluate our system in three scenarios with different lighting conditions by five users. The results show that UbiTouch achieves centimetre-level localization accuracy and poses no significant impact on the battery life. We envisage that UbiTouch could support applications such as text-writing and drawing.

Recently years, research in multi-robot systems has attracted increasingly attentions. One important research topic is to design programming models that can facilitate the developers to programme large-scale multi-robot systems. However, existing works fail to manage the robots to perform tasks with real-time requirements. To address this issue, we propose a new programming model called RMR (Real-time Multi-Robot). RMR is a logic programming model with real-time support. On the basis of the logic programming paradigm, RMR allows the code for multi-robot system to be written from a global perspective, rather than managing a large collection of independent robots. Moreover, RMR allows developers to set timing constraints on the behaviors of an ensemble of robots, which is not implemented by state of the art. After designing RMR, we further develop a compiler and a runtime system for distributed execution of RMR programs. To evaluate the performance of RMR, we deploy it in a simulator and a test-bed, and then demonstrate RMR based on several applications. Our results indicate that RMR greatly facilitates implementing correct collaborative multi-robot applications.

In complex terrain where mobile chargers hardly move around, a feasible solution to charge wireless sensor networks (WSNs) is using multiple fixed chargers to charge WSNs concurrently with relative long distance. Due to the radio interference in the concurrent charging, it is needed to schedule the chargers so as to facilitate each sensor node to harvest sufficient energy quickly. The challenge lies that each charger's charging utility cannot be calculated (or even defined) independently due to the nonlinear superposition charging effect caused by the radio interference. In this paper, we model the concurrent radio charging, and formulate the concurrent charging scheduling problem (CCSP) whose objective is to design a scheduling algorithm for the chargers so as to minimize the time spent on charging each sensor node with at least energy E. We prove that CCSP is NP-hard, and propose a greedy algorithm based on submodular set cover problem. We also propose a genetic algorithm for CCSP. Simulation results show that the performance of the greedy CCSP algorithm is comparable to that of the genetic algorithm.

Traffic lights in urban area can significantly influence the efficiency and effectiveness of transportation. The real-time scheduling information of traffic lights is fundamentally important for many intelligent transportation applications, such as shortest-time navigation and green driving advisory. However, existing traffic light scheduling identification systems either entail dedicated infrastructures or depend on specialized traffic traces, which hinders the popularity and real world deployment. Differently, we propose to identify real-time traffic light scheduling by analyzing taxi traces that are widely accessible from taxi companies. The key idea is to exploit the periodicity in traffic patterns, which is directly affected by traffic lights. We also develop advanced algorithms to identify red/green lights duration and signal change time. We evaluate our solution using over one billion taxi records from Shenzhen, China. The evaluation results validate the effectiveness of our system.

A software-defined network empowers the creation of a flexible network architecture by abstracting flow control from individual devices to the network level. In this paper, we address the challenges in applying SDN to develop high- performance vehicular networks. We present SDVN, a new SDN based vehicular network architecture. It organizes the topology of the vehicular networks and utilizes vehicle trajectory prediction to mitigate the overhead of the SDN control and data plane communication. Moreover, we propose a multicast protocol over SDVN, as multicast is the foundation of many vehicular network applications. The protocol exploits the network topology information provided by SDVN to make far more efficient multicast scheduling decision. The multicast scheduling problem is formulated to minimize the communication cost with bounded delay constraint. A polynomial time approximation algorithm is proposed. We conduct extensive experiments using traffic traces. The evaluation shows that the SDVN based multicast protocol outperforms existing decentralized approaches.

Wireless sensor networks (WSNs) have been considered to be the next generation paradigm of structural health monitoring (SHM) systems due to the low cost, high scalability and ease of deployment. Due to the intrinsically energy-intensive nature of the sensor nodes in SHM application, it is highly preferable that they can be divided into subsets and take turns to monitor the condition of a structure. This approach is generally called as `coverage-preserving scheduling' and has been widely adopted in existing WSN applications. The problem of partitioning the nodes into subsets is generally called as the 'maximum lifetime coverage problem (MLCP)'. However, existing solutions to the MLCP cannot be directly applied to SHM application. As compared to other WSN applications, we cannot define a specific coverage area independently for each sensor node in SHM, which is however the basic assumption in all existing solutions to the MLCP. In this paper, we proposed two approaches to solve the MLCP in SHM. The performance of the methods is demonstrated through both extensive simulations and real experiments.

Non-invasive human sensing based on radio signals has attracted numerous research interests in recent years. Previous work mainly focused on detecting the presence of a person or identifying human gestures and activities. In this paper, we show that with off-the-shelf WiFi devices, fine-grained respiration information of a person under different sleeping positions can be extracted successfully. We do this by introducing a breath monitoring system based on WiFi signals. This system adopts off-the-shelf WiFi devices to continuously collect the fine-grained wireless channel state information (CSI) around a person. From the CSI, the rhythmic patterns associated with respiration and abrupt changes due to the body movement are identified. Compared to existing respiration monitoring systems that usually require special devices attached to human body, this system is completely contactless. In addition, different from many vision-based sleep monitoring systems, it is robust to low-light environments and does not raise privacy concerns. Preliminary testing results show that our system can reliably track a person's respiration reliably in different sleeping postures.

Event-triggered wake-up, in which sensor nodes wake up to work in the presence of some pre-defined events, has been widely used in wireless sensor networks (WSNs) to save energy while still completing the tasks required. However, some recently emerged domain-specific WSN applications such as structural health monitoring (SHM) and volcano seismic tomography, have different requirements with regard to wake-up as compared to conventional WSN applications. In these domain-specific applications, the wake-up should be network-wide and nodes to be woken up are not limited to those close to event locations. In addition, the wake-up should be fast to capture enough information during generally short events and be reliable to avoid costly false-positive wake-ups. This problem has not been addressed in the literature. In this paper, we designed two types of wake-up units, based on which we propose a new chain-reaction wake-up mechanism to address this challenge. In this mechanism, we carefully select some nodes used to initiate the wake-up process, such that the wake-up delay is minimized under the false alarm constraint. We propose two greedy algorithms and a randomized one that leverages the solution to the classic Knapsack problem. The performance of the proposed wake-up mechanism is demonstrated through both simulation and experiments.

The potential of crowdsourcing for complex problem solving has been revealed by smartphones. Nowadays, vehicles have also been increasingly adopted as participants in crowd-sourcing applications. Different from smartphones, vehicles have the distinct advantage of predictable mobility, which brings new insight into improving the crowdsourcing quality. Unfortunately, utilizing the predictable mobility in participant recruitment poses a new challenge of considering not only current location but also the future trajectories of participants. Therefore, existing participant recruitment algorithms that only use the current location may not perform well. In this paper, based on the predicted trajectory, we present a new participant recruitment strategy for vehicle-based crowdsourcing. This strategy guarantees that the system can perform well using the currently recruited participants for a period of time in the future. The participant recruitment problem is proven to be NP-complete, and we propose two algorithms, a greedy approximation and a genetic algorithm, to find the solution for different application scenarios. The performance of our algorithms is demonstrated with traffic trace dataset. The results show that our algorithms outperform some existing approaches in terms of the crowdsourcing quality.

It is well recognized that walking while using mobile phones will make people more susceptible at various risks. Existing studies to improve smartphone users' safety are mainly limited to detecting incoming vehicles. They are not able to address some more common and equally dangerous accidents such as trips, falling from stairs, platforms or falling into an open manhole. These hazards are generally caused by sudden change of ground. In this paper, we propose UltraSee, the first system that is able to detect sudden change of ground for pedestrian mobile phone users. UltraSee augments smartphones with a small ultrasonic sensor which can detect the abrupt change of distance ahead. UltraSee also leverages the context information of smartphone usage such as screen status and holding orientation to improve detection accuracy and reduce energy consumption as well as unnecessary alarms. We have carried out extensive experiments in different scenarios and by different users. The results show that UltraSee can achieve accident detection rate of 94% with false positive rate of 4.4% and reduce unnecessary alarms by 90%. In terms of energy consumption, UltraSee costs only about 20% energy compared to the existing works that only rely on smartphone cameras.

Non-invasive human sensing based on radio signals has attracted a great deal of research interest and fostered a broad range of innovative applications of localization, gesture recognition, smart health-care, etc., for which a primary primitive is to detect human presence. Previous works have studied the detection of moving humans via signal variations caused by human movements. For stationary people, however, existing approaches often employ a prerequisite scenario-tailored calibration of channel profile in human-free environments. Based on in-depth understanding of human motion induced signal attenuation reflected by PHY layer channel state information (CSI), we propose DeMan, a unified scheme for non-invasive detection of moving and stationary human on commodity WiFi devices. DeMan takes advantage of both amplitude and phase information of CSI to detect moving targets. In addition, DeMan considers human breathing as an intrinsic indicator of stationary human presence and adopts sophisticated mechanisms to detect particular signal patterns caused by minute chest motions, which could be destroyed by significant whole-body motion or hidden by environmental noises. By doing this, DeMan is capable of simultaneously detecting moving and stationary people with only a small number of prior measurements for model parameter determination, yet without the cumbersome scenario-specific calibration. Extensive experimental evaluation in typical indoor environments validates the great performance of DeMan in various human poses and locations and diverse channel conditions. Particularly, DeMan provides a detection rate of around 95% for both moving and stationary people, while identifies human-free scenarios by 96%, all of which outperforms existing methods by about 30%.

Fault-tolerant event detection (FTED), whose objective is to correctly detect events of interest in the presence of faulty nodes, remains to be one of the hot research areas in wireless sensor networks (WSNs). However, many recently emerged `domain-specific' applications of WSNs, such as structural health monitoring (SHM) and volcano monitoring, have shown some distinct features from traditional WSN applications. For example, data collected each time from sensor nodes is not a scalar but a long dynamic data sequence. In addition, detecting an event in these applications generally requires low-level collaboration of multiple sensors. As a consequence, existing FTED schemes usually cannot work well in these applications. In this paper, we realize FTED in a typical domain-specific application of WSNs: SHM. The main contribution of this work is I-FUND, a faulty node detection algorithm that takes feature vectors as input and can even handle the `element mismatch problem' where comparable elements in vectors are located at unknown different positions. In addition, I-FUND adopts adaptive stop criterion identified from data and turns out to be reliable even when a large percentage of the sensor nodes report erroneous observations. The effectiveness of the proposed scheme is demonstrated through both simulations and real experiments.

Current researches on topology control with cooperative communication (CC) in wireless ad hoc networks have focused on network connectivity, path energy-efficiency and node transmission power reduction. However, fault-tolerance related issues have not been adequately addressed. In this paper, we propose a CC-based scheme to achieve more efficient fault-tolerant topology control. We first definek-connectivity under the CC model and then design a distributed scheme for building at-spanner withk-connectivity of an arbitrary communication network. Simulation results confirm that the proposed scheme can tolerate node failures as well as exploit the advantage of CC to achieve path energy-efficiency and lower power consumption of the network.

Due to the low cost and ease of deployment, wireless sensor networks (WSNs) are emerging as sensing paradigms that the structural engineering field has begun to consider as substitutes for traditional tethered structural health monitoring (SHM) systems. Different from other applications of WSNs such as environmental monitoring, SHM applications are much more data intensive and it is not feasible to stream the raw data back to the server due to the severe bandwidth and energy limitations of low-power sensor networks. In-network processing is a promising approach to address this problem but designing distributed versions for the sophisticated SHM algorithms is much more challenging because SHM algorithms are computationally intensive, and involve data-level collaboration of multiple sensors. In this paper, we select a classical SHM algorithm: the eigen-system realization algorithm (ERA), and propose a few distributed ERAs suitable for WSNs. In particular, we first design a method to incrementally calculate the ERA and then propose three schemes upon which the incremental ERA can be carried out along an Hamiltonian path, along a path in the minimum connected dominating set (MCDS) and along the shortest path tree (SPT). The efficacy of these schemes are demonstrated and compared through both simulation experiment. We believe the proposed schemes can also serve as a guideline when applying WSNs for other applications like SHM which are also data-intensive and involve sophisticated signal processing of collected information.

Co-existing in the unlicensed ISM band, ZigBee transmissions can be significantly interfered by WiFi. Although several approaches recently are proposed to enable ZigBee transmission under WiFi interference, the ZigBee throughput still decreases to zero when WiFi throughput (generated by D-ITG) is over 8Mbps. In this paper, we propose a real-time (<; 5ms) adaptive transmission (RAT) scheme to efficiently adapt forward error-correction coding (FEC) on ZigBee devices in dynamic WiFi environment. We find that sizes of WiFi frames well follow the power law distribution model. With the model, corruption in ZigBee packets can be estimated to some extent, thus facilitating ZigBee device to choose a suitable FEC coding to maximize the throughput. Extensive experimental results show that, compared with existing works, RAT achieves significant performance improvement of ZigBee transmissions in WiFi environment with different traffic load. Particularly, the ZigBee throughput of RAT can be about 10kbps when the WiFi throughput is 8Mbps.

Wireless sensor networks (WSNs) are generally used to monitor, in an area, certain phenomena which can be events or targets that users are interested. To extend the system lifetime, a widely used technique is `Energy-Efficient Coverage-Preserving Scheduling(EECPS)', in which at any time, only part of the nodes are activated to fulfill the function. To determine which nodes should be activated at a certain time is the key for the EECPS and this problem has been studied extensively. Existing solutions are based on the assumption that each node has a fixed coverage area, and once the event/target occurs in this area, it can be detected by this sensor. However, this coverage model is not always valid. In some applications such as structural health monitoring (SHM) and volcano monitoring, to fulfill a required function always requires low level collaboration from multiple sensors. The coverage area for individual sensor node therefore cannot be defined explicitly since single sensor is not able to fulfill the function alone, even it is close to the event or target to be monitored. In this paper, using an example of SHM, we illustrate how to support EECPS in some special applications of WSNs. We re-define the `coverage' and based on the new coverage model, two methods are proposed to partition the deployed sensor nodes into qualified cover sets such that the system lifetime can be maximized by letting these sets work by turns. The performance of the methods is demonstrated through extensive simulation and experiment.

Domain-specific applications, such as structural health monitoring, have been one of the main drivers that motivates the real-world deployment of wireless sensor networks. Due to their data-intensive nature, it is typical for these applications to make heavy uses of least squares estimation as a foundation for their algorithms, which is a standard approach to compute the approximate solution of sets of equations in which there are more equations than unknowns. Due to the very limited amount of energy and computation power available on the sensors, it is imperative to design new algorithms to perform least squares estimation in a distributed fashion. While we wish to conserving energy by minimizing communication with our design, constraints on communication delays will also need to be satisfied. In this paper, we propose E3, a new distributed algorithm specifically designed to guarantee the precision of least squares estimation in sensor networks, with the objective of minimizing the energy consumption incurred during communication, while observing constraints on application-specific communication delays. Compared to previous works, we show that E3maintains the same level of estimation precision while incurring much lower energy costs.

Vehicles can provide useful data to many urban computing applications. This paper addresses the issue of collecting data from multiple vehicles to a roadside base station using VANET. The impacts of real-time traffic condition have not been widely discussed in literature. In this paper, we study the data collection problem under different traffic conditions. The objective is to minimize the network communication overhead while satisfying the data collection time constraint. We formulate the problem as an scheduling optimization problem. A dynamic programming based solution and a genetic algorithm based solution are developed to solve the problem for different application scenarios. The solution can adaptively choose to carry or forward data based on current traffic information. Evaluation shows that the proposed solution outperforms some existing ones in terms of effectiveness and efficiency.

Is it possible to leverage WiFi signals collected in bedrooms to monitor a person's sleep? In this paper, we show that with off-the-shelf WiFi devices, fine-grained sleep information like a person's respiration, sleeping postures and rollovers can be successfully extracted. We do this by introducing Wi-Sleep, the first sleep monitoring system based on WiFi signals. Wi-Sleep adopts off-the-shelf WiFi devices to continuously collect the fine-grained wireless channel state information (CSI) around a person. From the CSI, Wi-Sleep extracts rhythmic patterns associated with respiration and abrupt changes due to the body movement. Compared to existing sleep monitoring systems that usually require special devices attached to human body (i.e. Probes, head belt, and wrist band), Wi-Sleep is completely contact less. In addition, different from many vision-based sleep monitoring systems, Wi-Sleep is robust to low-light environments and does not raise privacy concerns. Preliminary testing results show that the Wi-Sleep can reliably track a person's respiration and sleeping postures in different conditions.

An important category of mobile RFID systems is the RFID system with mobile RFID tags. The mobility of RFID tags poses new challenges to designing RFID anti-collision protocols. Existing RFID anti-collision protocols cannot support high tag moving speed and high identification rate simultaneously. These protocols do not distinguish the identification deadlines of moving tags. Also, when tags move fast, they cannot determine the number of unidentified tags in the interrogation area of an RFID reader. In this paper, we propose a schedule-based RFID anti-collision protocol which, given a high identification rate, achieves the maximal tag moving speed. The protocol, without the need to estimate the number of unidentified tags, schedules an optimal number of tags to compete for the channel according to their identification deadlines, so as to achieve the optimal identification performance. The simulation and experiment results show that our approach can increase the moving speed of tags significantly compared with existing approaches, while achieving a high identification rate.

Reliably detecting event in the presence of faulty nodes, particularly nodes with faulty readings is a fundamental task in wireless sensor networks (WSNs). Existing fault-tolerant event detection schemes usually 'mask' the effect of faulty readings through high-level fusion techniques. However, in some applications such as structural health monitoring (SHM) and volcano monitoring, detecting the events of interest requires lowlevel data collaboration from multiple sensors. This implies that the effect of faulty readings cannot be masked once they are involved into event detection. Nodes with faulty readings must be firstly detected and removed from the system. Unfortunately, most existing techniques to detect faulty nodes can only take boolean or scalar data as input while in these applications, data generated from each sensor is a sequence of dynamic data. In this paper, we address these issues using an example of SHM. Detecting event in SHM (i.e. structural damage) requires low level collaboration from multiple sensors, and each sensor generates a sequence of dynamic vibrational data. We proposed a fault-tolerant event detection scheme in SHM called FTED. In FTED, three novel techniques are proposed: (1) distributed extraction of features for faulty node detection, (2) iterative faulty node detection (I-FUND), and (3) distributed event detection. In particular, I-FUND takes vector as input and can even handle the 'element mismatch problem' where comparable elements in vectors are located at unknown different positions. The effectiveness of FTED is demonstrated through both simulations and real experiments.

Chirp Spread Spectrum (CSS) based localization techniques are becoming more attractive as they provide improved localization accuracy and robustness compared to WiFi or ZigBee based approaches. However, a remaining problem is the necessary update frequency: In existing CSS based localization systems, the positions of the objects are determined one by one via unicast with nearby anchors instead of using broadcasts. We propose a faster distributed localization scheme for CSS based systems. A portion of nodes are considered cluster heads; they determine the locations of un-localized nodes by dynamically increasing the transmission power. Our novel scheme not only fully utilizes the spatial redundancy, which is crucial for speeding up the localization process. By also allowing to establish new anchors in a two-hop range, we can further increase speed without significantly influencing localization error. The performance of the proposed method is demonstrated through simulation.

Information propagation in delay tolerant networks (DTN) is difficult due to the lack of continues connectivity. Most of previous work put their focus on the information propagation in static network. In this work, we examine two closely related problems on information propagation in predicable DTN. In particular, we assume that during a certain time period, the interacting process among nodes is known a priori or can be predicted. The first problem is to select a set of initial source nodes, subject to budget constraint, in order to maximize the total weight of nodes that receive the information at the final stage. This problem is well-known influence maximization problem which has been extensively studied for static networks. The second problem we want to study is minimum cost initial set problem, in this problem, we aim to select a set of source nodes with minimum cost such that all the other nodes can receive the information with high probability. We conduct extensive experiments using $10,000$ users from real contact trace.

Event-triggered wake-up, in which sensor nodes wake up to work in the presence of some predefined events, has been widely used in wireless sensor networks (WSNs) to save energy while still fulfilling the tasks required. However, existing mechanisms for event-triggered wake-up, such as using auxiliary low-power sensors or via wireless radio communication, cannot be applied to some domain-specific applications such as structural health monitoring (SHM) and volcano monitoring(VM). These applications have different requirements about the wake-up from 'conventional' WSN applications. Firstly, the wake-up should be network-wide and nodes to be woken up are not limited to those close to event location. In addition, the wake-up should be reliable to avoid unnecessary false wake-ups. Finally, the wake-up should be fast to capture the short events in these applications. How to realize network-wide, reliable and fast wake-up in a WSN is a challenging task that has not been addressed in literature. In this paper, based on the developed two types of wake-up units, we designed a novel chain-reaction wake-up mechanism to fulfill the task. As the key in this mechanism, we identify the sentry node placement problem in which some nodes used to initialize the wake-up process are carefully selected such that the wake-up delay is minimized under the false alarm constraint. We propose two greedy algorithms and a randomized one which leverages the solution to the classic Knapsack problem. The performance of the proposed wake-up mechanism is demonstrated through both simulation and experiments.

In recent years, there has been an increasing interest in the adoption of wireless sensor networks (WSNs) for structural health monitoring (SHM). However, considering the large amount of sampled data, limited power supply and wireless bandwidth of WSNs, it is generally not possible for sensor nodes to monitor structural condition continuously especially for long-term SHM. From SHM perspective, it is highly desirable to collect data during the occurrence of some certain kinds of events such as earthquakes, large wind, etc, since data collected during these periods are more informative for damage detection purpose. However, these events in SHM occur infrequently and if happen, only last for a very short period of time. To effectively capture these short events using energy-limited wireless sensor nodes is a challenging task that has not been addressed in literature. In this paper, we propose a fast and reliable scheme to capture these short-term events. In terms of hardware, the radio-triggered unit and the vibration-triggered unit are designed in our motes. From software perspective, we develop the event capture scheme to realize fast, reliable and energy-efficient event detection under noisy environment. We mainly study the very first problem of the scheme: sentry selection. The optimization problem is formulated and we show that the problem is a general case of the classical k-center problem. Then we propose a greedy algorithm to solve this problem, and conduct simulation to test the effectiveness of the proposed algorithm.

To ensure sustainable operations of wireless sensor networks, environmental energy harvesting has been well recognized as one promising solution for long-term applications. Unlike in battery-powered sensor networks, we are targeting a duty-cycle adjustment to optimize the network performance, e.g., delay minimization, with full harvested energy utilization. In this paper, we introduce a set of duty-cycle adjustment schemes that will minimize cross traffic delay (CTD) in energy-harvesting sensor networks. We first present an offline solution by assuming that the link reliability and traffic distribution are known a priori. Based on the submodular property of the CTD function, we theoretically prove that a simple greedy algorithm can achieve constant approximation. We next propose a class of online algorithms that do not require the knowledge of link reliability and traffic distribution. For each of these algorithms, we give a theoretical bound on the performance. We have evaluated our design with a TelosB-based implementation and experimental results corroborate our theoretical analysis.

In many applications of WSNs, the events occur infrequently but once they occur, the corresponding information needs to be sent to the sink node in a short period of time for the necessary reactions. Detection of rare events in a fast and energy-efficient manner is an important issue in WSNs. In this paper, based on the optimal solution for the Best-choice Problem with Bounded Random Observation Number, we propose an Energy-efficient Online Decision Scheme (EODS) to handle this problem. Combining with the design of nodes' duty cycle, the EODS avoids the redundant transmissions and achieves the tradeoff between delay and energy efficiency. Simulation results reveal that the EODS achieves a good balance between delay and energy efficiency.

In recent years, using wireless sensor networks (WSNs) for structural health monitoring (SHM) has attracted increasing attention. Traditional centralized SHM algorithms developed by civil engineers can achieve the highest damage detection quality since they have the raw data from all the sensor nodes. However, directly implementing these algorithms in a typical WSN is impractical considering the large amount of data transmissions and extensive computations required. Correspondingly, many SHM algorithms have been tailored for WSNs to become distributed and less complicated. However, the modified algorithms usually cannot achieve the same damage detection quality of the original centralized counterparts. In this paper, we select a classical SHM algorithm: the eigen-system realization algorithm (ERA), and propose a distributed version for WSNs. In this approach, the required computations in the ERA are updated incrementally along a path constructed from the deployed sensor nodes. This distributed version is able to achieve the same quality of the original ERA using much smaller wireless transmissions and computations. The efficacy of the proposed approach is demonstrated through both simulation and experiment.

Wireless sensor networks (WSNs) have become an increasingly compelling platform for structural health monitoring (SHM) due to relatively low-cost, easy installation, etc. However, the challenge of effectively monitoring structural health condition (e.g., damage) under WSN constraints (e.g., limited energy, narrow bandwidth) and sensor faults has not been studied before. In this paper, we focus on tolerating sensor faults in WSN-based SHM. We design a distributed WSN framework for SHM and then examine its ability to cope with sensor faults. We bring attention to an undiscovered yet interesting fact, i.e., the real measured signals introduced by faulty sensors may cause an undamaged location to be identified as damaged (false positive) or a damaged location as undamaged (false negative) diagnosis. This can be caused by faults in sensor bonding, precision degradation, amplification gain, bias, drift, noise, and so forth. We present a distributed algorithm to detect such types of faults, and offer an online signal reconstruction algorithm to recover from the wrong diagnosis. Through simulations and a WSN prototype system, we evaluate the effectiveness of our proposed algorithms.

Fault tolerance in wireless sensor networks (WSNs) has been studied extensively by computer science researchers and they proposed many fault-tolerant schemes for various applications including target and event detection. However, these schemes would fail in a particular application of WSNs: structural health monitoring (SHM). Different from other applications of WSNs, detecting structural damage requires significant amount of civil domain knowledge and utilizes different detection model. Meanwhile, researchers in civil engineering also proposed some fault-tolerant SHM algorithms. However, these algorithms are all centralized and not applicable to resource-limited wireless sensor networks. To our best knowledge, we are the first to address fault tolerance problem in WSN-based SHM. We target faulty sensor reading, one of the most difficult types of sensor fault to be detected, and propose a fault-tolerant SHM approach. The proposed approach is lightweight and it is able to disambiguate structural damage from sensor faults. The effectiveness of the proposed approach is demonstrated through both simulation and real implementation.

In most of the existing video surveillance systems, captured video streams by CCTV or video cameras are aggregated through cables to a central station and monitored by associated human operators. However, the high deployment cost, low scalability and reuseability still remain to be the main roadblock for their wider application. Using human to monitor events can also be unreliable. Accordingly, we designed a ubiquitous wireless video surveillance system. This system uses wireless sensor nodes to detect pre-defined events and utilizes wireless mesh network to transmit high-quality video streams. Without cables, the deployment cost is significantly decreased. Different application users can also define various events and automatically receive notification and corresponding video streams when the events occur. In addition, mobile users, even when roaming, can access this system. In this demo, we introduce the hardware and software of this system and its implementation in our intelligent transportation system testbed.

In most of the applications of wireless sensor networks(WSN), covering the area of interest and delivering the sensed information to the sink are two fundamental functions. Extensive research associated with these two issues, such as energy efficient coverage and delay-constraint routing, can be found in the literature. However, few works combine these two issues together. Considering the fact that wireless sensors can take the responsibility of both sensing and routing, it is expected that a solution jointly considering these two issues will provide more benefit. In this paper, we consider the problem: how to find a routing path in a WSN with the maximum sensing coverage provided by the nodes on the path subject to the delay constraint. We first proved that this problem is NP-hard and then proposed a distributed algorithm based on Monte-Carlo integration method and label setting(LS) algorithm. Analysis and simulation results show that under the same time delay constraint, the proposed algorithm can find a routing path with significantly larger sensing coverage (more than $87\%$ in our simulation) than that was obtained considering only hop constraint.

In recent years, research on using wireless sensor networks (WSNs) for structural health monitoring (SHM) has attracted increasing attention. Unlike other monitoring applications, detection of possible structure damage requires significant amount of domain knowledge that computer science researchers are usually unfamiliar with. As a result, most previous work in WSN-based SHM was done by researchers in civil engineering. However, civil researchers often tend to solve practical engineering problems but rarely consider designing a system in an optimal way, particularly when the limited wireless bandwidth and restricted resources of WSNs need to be addressed. Through the collaboration with civil researchers, we demonstrate that optimization design can significantly help improve the performance of a WSN-based SHM system. We consider a fundamental problem in SHM: modal analysis, which is used to obtain the dynamic structural vibration characteristics. Cluster-based modal analysis approach is adopted. In each cluster, the vibration characteristics are identified and then are assembled together. Different from other applications, clustering in this approach should meet some extra requirements of modal analysis. Moreover, cluster size should be optimized to minimize the total energy consumption. This clustering problem is formally formulated and proven to be NP complete. Two centralized and one distributed algorithms are proposed to solve the problem. The effectiveness and efficiency of the proposed cluster-based modal analysis along with the clustering algorithms are evaluated using both simulation and experiments.

In this work, we design efficient algorithms for scheduling node activities, under the physical interference model, to minimize the delay for activating a set of communication links, or for finishing a data aggregation communication task. Given a set of communication links, assume that each link is associated with a positive weight (representing the award of transmission along this link). We consider two problems: the first one is to find an independent set of links with maximum total weight; the second one is to partition all links into independent subsets, such that the number of subsets is minimized. We are the first to develop distributed algorithms with constant approximations for both problems respectively. The other line of this work is to explore the relations between link scheduling and an important practical problem: Minimum Latency Aggregation Scheduling which seeks a shortest schedule for data aggregation in multi-hop wireless networks. By utilizing the algorithmic results for link scheduling, our proposed method can find an aggregation schedule that greatly improves the upper bound on latency, compared to the previous best result.

Most researchers encountered the following two problems when working with real Wireless Sensor Networks (WSNs): (1) Sensor nodes cannot satisfy application requirements even though the nominal sensing/transmission rates of these nodes are much higher than required. (2) In a WSN deployed in a large area, it is difficult or infeasible to get runtime performance evaluation of sensor nodes. We found out that the root reason of these two problems is resource competition, in which an operation has to wait for the resources being used by other operations. Therefore, we propose a dual-mote testbed, which is able to avoid both hardware competition on a node and wireless channel competition in a network. We have implemented the hardware, supporting protocols and tools of the testbed. The experimental results show that, compared to a general WSN, the improvement on performances such as throughput and response speed by our testbed are more than doubled.