Search

Personalized location prediction is key to many mobile applications and services. In this paper, motivated by both statistical and visualized preliminary analysis on three real datasets, we observe a strong spatiotemporal correlation for user trajectories among the visited area-of-interests (AoIs) and different time periods on both weekly and daily basis, which directly motivates our time-aware location prediction model design called ``$t$-LocPred". It models the spatial correlations among AoIs by coarse-grained convolutional processing of the user trajectories in AoIs of different time periods (``ConvAoI"); and predicts his/her fine-grained next visited PoI using a novel memory-augmented attentive LSTM model ("mem-attLSTM") to capture long-term behavior patterns. Experimental results show that t-LocPred outperforms 8 baselines. We also show the impact of hyperparameters and the benefits ConvAoI can bring to these baselines.

Online social networks, such as Facebook and Twitter, enable users to share their shopping/travel experiences with their friends. However the influence on users’ decision-making on next visit/buy has sparse research exposure, by accurately modeling long-term user behaviors from historical data. The existing methods do not fully take advantage of the underlying social networks to model user interests, nor they have not modeled long-term transitional behavior patterns. In this paper, we propose a novel Social Influence aware and Memory augmented Sequence learning (SIMS) model, on what a user will likely buy/visit next. Specifically, SIMS first learns a representation for the visiting/purchasing sequence of each user using the sequence-to-sequence learning method. Then it predicts the interest of a user by integrating the representation of his/her own sequence, with another representation of the corresponding social influence, which is learned using an autoencoder-based model. In addition, we leverage an emerging memory augmented neural network, Differentiable Neural Computer (DNC), to further improve prediction accuracy. We conduct extensive experiments to evaluate the proposed model using three real-world datasets, Yelp, Epinions and Ciao. When compared with 10 other baselines and state-of-the-art solutions, the experimental results show that 1) the proposed model significantly outperforms all other methods in terms of various accuracy-related metrics; 2) the proposed social influence modeling and memory augmentation do lead to the performance gain.

Tremendous research efforts have been made to thrive deep domain adaptation (DA) by seeking domain-invariant features. Most existing deep DA models only focus on aligning feature representations of task-specific layers across domains while integrating a totally shared convolutional architecture for source and target. However, we argue that such strongly-shared convolutional layers might be harmful for domain-specific feature learning when source and target data distribution differs to a large extent. In this paper, we relax a shared-convnets assumption made by previous DA methods and propose a Domain Conditioned Adaptation Network (DCAN), which aims to excite distinct convolutional channels with a domain conditioned channel attention mechanism. As a result, the critical low-level domain-dependent knowledge could be explored appropriately. As far as we know, this is the first work to explore the domain-wise convolutional channel activation for deep DA networks. Moreover, to effectively align high-level feature distributions across two domains, we further deploy domain conditioned feature correction blocks after task-specific layers, which will explicitly correct the domain discrepancy. Extensive experiments on three cross-domain benchmarks demonstrate the proposed approach outperforms existing methods by a large margin, especially on very tough cross-domain learning tasks., Accepted by AAAI 2020

In this paper, we propose to leverage the emerging deep learning techniques for sequential modeling of user interests based on big social data, which takes into account influence of their social circles. First, we present a preliminary analysis for two popular big datasets from Yelp and Epinions. We show statistically sequential actions of all users and their friends, and discover both temporal autocorrelation and social influence on decision making, which motivates our design. Then, we present a novel hybrid deep learning model, Social-Aware Long Short-Term Memory (SA-LSTM), for predicting the types of item/PoIs that a user will likely buy/visit next, which features stacked LSTMs for sequential modeling and an autoencoder-based deep model for social influence modeling. Moreover, we show that SA-LSTM supports end-to-end training. We conducted extensive experiments for performance evaluation using the two real datasets from Yelp and Epinions. The experimental results show that (1) the proposed deep model significantly improves prediction accuracy compared to widely used baseline methods; (2) the proposed social influence model works effectively; and (3) going deep does help improve prediction accuracy but a not-so-deep deep structure leads to the best performance.

Domain adaptation (DA) enables knowledge transfer from a labeled source domain to an unlabeled target domain by reducing the cross-domain distribution discrepancy. Most prior DA approaches leverage complicated and powerful deep neural networks to improve the adaptation capacity and have shown remarkable success. However, they may have a lack of applicability to real-world situations such as real-time interaction, where low target inference latency is an essential requirement under limited computational budget. In this paper, we tackle the problem by proposing a dynamic domain adaptation (DDA) framework, which can simultaneously achieve efficient target inference in low-resource scenarios and inherit the favorable cross-domain generalization brought by DA. In contrast to static models, as a simple yet generic method, DDA can integrate various domain confusion constraints into any typical adaptive network, where multiple intermediate classifiers can be equipped to infer "easier" and "harder" target data dynamically. Moreover, we present two novel strategies to further boost the adaptation performance of multiple prediction exits: 1) a confidence score learning strategy to derive accurate target pseudo labels by fully exploring the prediction consistency of different classifiers; 2) a class-balanced self-training strategy to explicitly adapt multi-stage classifiers from source to target without losing prediction diversity. Extensive experiments on multiple benchmarks are conducted to verify that DDA can consistently improve the adaptation performance and accelerate target inference under domain shift and limited resources scenarios, Accepted by CVPR 2021

Understanding the movement patterns of humans and vehicles traveling in a city is important for many applications like emergency evacuation and rescue, as well as city planning and management. In this paper, we aim to predict citywide crowd flows within a period in the future to give aid to urban management, through modeling spatiotemporal patterns of recent crowd flows. We present a novel deep model for this task, called "AttConvLSTM", which leverages a convolutional LSTM (ConvLSTM), Convolutional Neural Networks (CNNs) along with an attention mechanism, where ConvLSTM keeps spatial information as intact as possible during sequential analysis, and the attention mechanism can focus important crowd flow variations which cannot be identified by the recurrent module. We conducted extensive experiments for performance evaluation using three large datasets, including Beijing Taxi dataset, Rome Taxi dataset, and Chengdu Didi chauffeuring trace. The experimental results show that AttConvLSTM significantly outperforms several widely-used baselines in terms of Root Mean Squared Error (RMSE), and Mean Average Percentage Error (MAPE), indicating that our approach can deal with crowd flows with different dynamics in both spatial and temporal domains, and make valid predictions several steps ahead.

Real-world training data usually exhibits long-tailed distribution, where several majority classes have a significantly larger number of samples than the remaining minority classes. This imbalance degrades the performance of typical supervised learning algorithms designed for balanced training sets. In this paper, we address this issue by augmenting minority classes with a recently proposed implicit semantic data augmentation (ISDA) algorithm, which produces diversified augmented samples by translating deep features along many semantically meaningful directions. Importantly, given that ISDA estimates the class-conditional statistics to obtain semantic directions, we find it ineffective to do this on minority classes due to the insufficient training data. To this end, we propose a novel approach to learn transformed semantic directions with meta-learning automatically. In specific, the augmentation strategy during training is dynamically optimized, aiming to minimize the loss on a small balanced validation set, which is approximated via a meta update step. Extensive empirical results on CIFAR-LT-10/100, ImageNet-LT, and iNaturalist 2017/2018 validate the effectiveness of our method., Comment: Accepted at CVPR 2021

Domain adaptation has been widely explored by transferring the knowledge from a label-rich source domain to a related but unlabeled target domain. Most existing domain adaptation algorithms attend to adapting feature representations across two domains with the guidance of a shared source-supervised classifier. However, such classifier limits the generalization ability towards unlabeled target recognition. To remedy this, we propose a Transferable Semantic Augmentation (TSA) approach to enhance the classifier adaptation ability through implicitly generating source features towards target semantics. Specifically, TSA is inspired by the fact that deep feature transformation towards a certain direction can be represented as meaningful semantic altering in the original input space. Thus, source features can be augmented to effectively equip with target semantics to train a more transferable classifier. To achieve this, for each class, we first use the inter-domain feature mean difference and target intra-class feature covariance to construct a multivariate normal distribution. Then we augment source features with random directions sampled from the distribution class-wisely. Interestingly, such source augmentation is implicitly implemented through an expected transferable cross-entropy loss over the augmented source distribution, where an upper bound of the expected loss is derived and minimized, introducing negligible computational overhead. As a light-weight and general technique, TSA can be easily plugged into various domain adaptation methods, bringing remarkable improvements. Comprehensive experiments on cross-domain benchmarks validate the efficacy of TSA., Comment: Accepted as CVPR 2021. The code is publicly available at https://github.com/BIT-DA/TSA

Deep learning (DL) models are increasingly built on federated edge participants holding local data. To enable insight extractions without the risk of information leakage, DL training is usually combined with differential privacy (DP). The core theme is to tradeoff learning accuracy by adding statistically calibrated noises, particularly to local gradients of edge learners, during model training. However, this privacy guarantee unfortunately degrades model accuracy due to edge learners' local noises, and the global noise aggregated at the central server. Existing DP frameworks for edge focus on local noise calibration via gradient clipping techniques, overlooking the heterogeneity and dynamic changes of local gradients, and their aggregated impact on accuracy. In this article, we present a systematical analysis that unveils the influential factors capable of mitigating local and aggregated noises, and design PrivateDL to leverage these factors in noise calibration so as to improve model accuracy while fulfilling privacy guarantee. PrivateDL features on: (i) sampling-based sensitivity estimation for local noise calibration and (ii) combining large batch sizes and critical data identification in global training. We implement PrivateDL on the popular Laplace/Gaussian DP mechanisms and demonstrate its effectiveness using Intel BigDL workloads, i.e., considerably improving model accuracy by up to 5X when comparing against existing DP frameworks.

Crowd saliency prediction refers to predicting where people look at in crowd scene. Humans have remarkable ability to rapidly direct their gaze to select visual information of interest when looking at a visual scene. Until now, research efforts are still focused on what type of feature is representative for crowd saliency, and which type of learning model is robust for crowd saliency prediction. In this paper, we propose a Random Forest (RF) based crowd saliency prediction approach with optimal feature combination, i.e., the Feature Combination Selection for Crowd Saliency (FCSCS) framework. More specifically, we first define three representative crowd saliency features, namely, FaceSizeDiff, FacePoseDiff and FaceWhrDiff. Next, we adopt the Random Forest (RF) algorithm to construct our saliency learning model. Then, we evaluate the performance of FCSCS framework with different feature combinations (fifteen combinations in our experiments). Those selected features include low-level features (i.e., color, intensity, orientation), four crowd features (i.e., face size, face density, frontal face, profile face) and three new defined features (i.e., FaceSizeDiff, FacePoseDiff and FaceWhrDiff). We use FCSCS framework to obtain the optimal feature combination that is most suitable for crowd saliency prediction and further train the saliency model based on the optimal feature combination. After that, we evaluate the performance of the crowd saliency prediction classifiers. Finally, we conduct extensive experiments and empirical evaluation to demonstrate the satisfactory performance of our approach.

Domain adaptation aims to transfer the enriched label knowledge from large amounts of source data to unlabeled target data. It has raised significant interest in multimedia analysis. Existing researches mainly focus on learning domain-wise transferable representations via statistical moment matching or adversarial adaptation techniques, while ignoring the class-wise mismatch across domains, resulting in inaccurate distribution alignment. To address this issue, we propose a Joint Adversarial Domain Adaptation (JADA) approach to simultaneously align domain-wise and class-wise distributions across source and target in a unified adversarial learning process. Specifically, JADA attempts to solve two complementary minimax problems jointly. The feature generator aims to not only fool the well-trained domain discriminator to learn domain-invariant features, but also minimize the disagreement between two distinct task-specific classifiers' predictions to synthesize target features near the support of source class-wisely. As a result, the learned transferable features will be equipped with more discriminative structures, and effectively avoid mode collapse. Additionally, JADA enables an efficient end-to-end training manner via a simple back-propagation scheme. Extensive experiments on several real-world cross-domain benchmarks, including VisDA-2017, ImageCLEF, Office-31 and digits, verify that JADA can gain remarkable improvements over other state-of-the-art deep domain adaptation approaches.

The core of many large-scale machine learning (ML) applications, such as neural networks (NN), support vector machine (SVM), and convolutional neural network (CNN), is the training algorithm that iteratively updates model parameters by processing massive datasets. From a plethora of studies aiming at accelerating ML, being data parallelization and parameter server, the prevalent assumption is that all data points are equivalently relevant to model parameter updating. In this article, we challenge this assumption by proposing a criterion to measure a data point's effect on model parameter updating, and experimentally demonstrate that the majority of data points are non-critical in the training process. We develop a slim learning framework, termed SlimML, which trains the ML models only on the critical data and thus significantly improves training performance. To such an end, SlimML efficiently leverages a small number of aggregated data points per iteration to approximate the criticalness of original input data instances. The proposed approach can be used by changing a few lines of code in a standard stochastic gradient descent (SGD) procedure, and we demonstrate experimentally, on NN regression, SVM classification, and CNN training, that for large datasets, it accelerates model training process by an average of 3.61 times while only incurring accuracy losses of 0.37 percent.

Crowd sensing campaigns encourage ordinary people to collect and share sensing data by using their carried smart devices. However, new challenges that must be faced have arisen. One of them is how to allocate tasks to the most appropriate participants when considering their different incentive requirements and credibility, in order to best satisfy the quality-of-information (QoI) requirements of multiple concurrent tasks, with different, and limited budget constraints. Another challenge is how to maximize participants’ rewards to encourage them to contribute sensing data continuously. To this end, in this paper, we first propose a crowd sensing system, that aims to address the above two challenges, where the system considers the benefits of both platform and participants. Then, a participant reputation definition and update method is proposed, that takes participant’s willingness and contributed data quality into consideration. Last, we introduce two metrics called “QoI satisfaction” and “Difficulty of Task (DoT)”. The former quantifies how much collected sensing data can satisfy the multi-dimensional task’s QoI requirements in terms of data quality, granularity and quantity, and the later aids participants to choose proper tasks to maximize their rewards. Finally, we compare our proposed scheme with existing methods via extensive simulations based on a real dataset. Extensive simulation results well justify the effectiveness and robustness of our approach.

Mobile crowdsourcing systems are important sources of information for the Internet of Things (IoT) such as gathering location related sensing data for various applications by employing ordinary citizens to participate in data collection. In order to improve the Quality of Information (QoI) of the collected data, the system server needs to coordinate participants with different data collection capabilities and various incentive requirements. However, existing participant coordination methods require the participants to reveal their trajectories to the system server which causes privacy leakage. But, with the improvement of ordinary citizens’ consciousness to protect their rights, the risk of privacy leakage may reduce their enthusiasm for data collection. In this paper, we propose a participant coordination framework, which allows the system server to provide optimal QoI for sensing tasks without knowing the trajectories of participants. The participants work cooperatively to coordinate their sensing tasks instead of relying on the traditional centralized server. A cooperative data aggregation, an incentive distribution method, and a punishment mechanism are further proposed to both protect participant privacy and ensure the QoI of the collected data. Simulation results show that our proposed method can efficiently select appropriate participants to achieve better QoI than other methods, and can protect each participant’s privacy effectively.

Modern communication networks have become very complicated and highly dynamic, which makes them hard to model, predict and control. In this paper, we develop a novel experience-driven approach that can learn to well control a communication network from its own experience rather than an accurate mathematical model, just as a human learns a new skill (such as driving, swimming, etc). Specifically, we, for the first time, propose to leverage emerging Deep Reinforcement Learning (DRL) for enabling model-free control in communication networks; and present a novel and highly effective DRL-based control framework, DRL-TE, for a fundamental networking problem: Traffic Engineering (TE). The proposed framework maximizes a widely-used utility function by jointly learning network environment and its dynamics, and making decisions under the guidance of powerful Deep Neural Networks (DNNs). We propose two new techniques, TE-aware exploration and actor-critic-based prioritized experience replay, to optimize the general DRL framework particularly for TE. To validate and evaluate the proposed framework, we implemented it in ns-3, and tested it comprehensively with both representative and randomly generated network topologies. Extensive packet-level simulation results show that 1) compared to several widely-used baseline methods, DRL-TE significantly reduces end-to-end delay and consistently improves the network utility, while offering better or comparable throughput; 2) DRL-TE is robust to network changes; and 3) DRL-TE consistently outperforms a state-ofthe-art DRL method (for continuous control), Deep Deterministic Policy Gradient (DDPG), which, however, does not offer satisfying performance., 9 pages, 12 figures, paper is accepted as a conference paper at IEEE Infocom 2018

The evolution of smartphones together with increasing computational power has empowered developers to create innovative context-aware applications for recognizing user-related social and cognitive activities in any situation and at any location. The existence and awareness of the context provide the capability of being conscious of physical environments or situations around mobile device users. This allows network services to respond proactively and intelligently based on such awareness. The key idea behind context-aware applications is to encourage users to collect, analyze, and share local sensory knowledge in the purpose for a large-scale community use by creating a smart network. The desired network is capable of making autonomous logical decisions to actuate environmental objects and also assist individuals. However, many open challenges remain, which are mostly arisen because the middleware services provided in mobile devices have limited resources in terms of power, memory, and bandwidth. Thus, it becomes critically important to study how the drawbacks can be elaborated and resolved and, at the same time, better understand the opportunities for the research community to contribute to the context-awareness. To this end, this paper surveys the literature over the period of 1991–2014 from the emerging concepts to applications of context-awareness in mobile platforms by providing up-to-date research and future research directions. Moreover, it points out the challenges faced in this regard and enlightens them by proposing possible solutions.

Roadside units (RSUs) are public and personal wireless access points that can provide communications with infrastructure in ad hoc vehicular networks. We present CLOCS (Counting and Localization using Online Compressive Sensing), a novel system to retrieve both the number and locations of RSUs through wardriving. CLOCS employs online compressive sensing (CS), where received signal strength (RSS) values are recorded at runtime, and the number and location of RSUs are recovered immediately based on limited RSS readings. CLOCS also uses fine retrieval based on an expectation maximization method along the driving route. Extensive simulation results and experiments in a real testbed deployed in the campus of the University of California, Irvine, confirm that CLOCS can successfully reduce the number of measurements for RSU recovery, while maintaining satisfactory counting and localization accuracy. In addition, data dissemination, time cost, and effects of different mobile scenarios using CLOCS are analyzed, and the impact of CLOCS on network connectivity is studied using Microsoft VanLan traces.

Today's smartphones are also fundamentally transforming the traditional understanding of “crowdsourcing” to an emerging type of participatory task-oriented application. It aims to support the so-called “Citizen Science efforts” for knowledge discovery to understand human behavior and measure/evaluate their opinions. To facilitate these scenarios, in this paper, we propose a novel efficient network management framework for participatory crowdsourcing. Specifically, we first formulate the optimization problem and propose a closed-form optimal solution to meet the quality-of-information (QoI) requirements of the task while minimizing the energy consumption variance among participants. We then largely extend the traditional framework of Gur Game for distributed decision making to recommend different levels of information contribution for each participant by merging multiple automaton chains into a single chain with multiple steady states. By modeling the user bidding behaviors, we propose a few incentive-based participant selection schemes to maximize the platform's benefits and meet participants' expectations. We extensively evaluate the proposed schemes under the Massachusetts Institute of Technology (MIT) Social Evolution data set, where both QoI requirements of the request and credit saving are successfully achieved, with a satisfactory level of energy consumption fairness among participants.

In existing wireless networks, supporting multimedia services are becoming more popular and important. In general, wireless multimedia networks should require energy efficiency and reliable transmission while keeping satisfactory quality of services. In this respect, cooperative communications have been considered as an efficient approach to address these demands by offering significant diversity gains over single-antenna systems without increasing requirements on radio resources. In this paper, we propose a power-allocation method to optimize the decode-and-forward (DF) cooperative transmission for source and relay nodes as a means to reduce the total power consumption, while maintaining the required quality of services, and investigate fundamental characteristics of cooperative transmission in terms of power efficiency. Moreover, for a network with multiple cooperative nodes, we also propose an energy-efficient relay selection rule to offer fairness at each node and implement it into a practical routing protocol. Our performance analysis is supplemented by simulation results to illustrate significant energy savings of the proposed methods.

Participatory sensing is now becoming more popular and has shown its great potential in various applications. It was originally proposed to recruit ordinary citizens to collect and share massive amounts of sensory data using their portable smart devices. By attracting participants and paying rewards as a return, incentive mechanisms play an important role to guarantee a stable scale of participants and to improve the accuracy/coverage/timeliness of the sensing results. Along this direction, a considerable amount of research activities have been conducted recently, ranging from experimental studies to theoretical solutions and practical applications, aiming at providing more comprehensive incentive procedures and/or protecting benefits of different system stakeholders. To this end, this paper surveys the literature over the period of 2004–2014 from the state of the art of theoretical frameworks, applications and system implementations, and experimental studies of the incentive strategies used in participatory sensing by providing up-to-date research in the literature. We also point out future directions of incentive strategies used in participatory sensing.

Mobile Crowd sensing is a new paradigm that encourages ordinary people to collect and share sensing data with their smart devices. However, the uncontrollable data quality is one of the critical problems that is potentially harmful to guarantee the availability and preciseness of mobile crowd sensing based services. In this paper, we propose a quality aware data collection mechanism based on a realistic scenario where participants arrive and report their conditions sequentially one by one. When a participant arrives, the mechanism first forecasts the amount of high quality data he#x002F;she may contribute by employing the expectation of Binomial-Poisson distribution, and then a two-level iterative algorithm is employed to calculate its parametric values. After, our designed mechanism decides to select the participant or not by combining with his#x002F;her requested reward. Extensive simulation results well justify the effectiveness and robustness of our approach, compared with another schemes.

By using increasingly popular smartphones, participatory sensing systems can collect comprehensive sensory data to retrieve context-aware information for different applications (or sensing tasks). However, new challenges arise when selecting the most appropriate participants when considering their different incentive requirements, associated sensing capabilities, and uncontrollable mobility, to best satisfy the quality-of-information (QoI) requirements of multiple concurrent tasks with different budget constraints. This paper proposes a multitask-oriented participant selection strategy called “DPS,” which is used to tackle the aforementioned challenges, where three key design elements are proposed. First is the QoI satisfaction metric, where the required QoI metrics of the collected data are quantified in terms of data granularity and quantity. Second is the multitask-orientated QoI optimization problem for participant selection, where task budgets are treated as the constraint, and the goal is to select a minimum subset of participants to best provide the QoI satisfaction metrics for all tasks. The optimization problem is then converted to a nonlinear knapsack problem and is solved by our proposed dynamic participant selection (DPS) strategy. Third is how to compute the expected amount of collected data by all (candidate) participants, where a probability-based movement model is proposed to facilitate such computation. Real and extensive trace-based simulations show that, given the same budget, the proposed participant selection strategy can achieve far better QoI satisfactions for all tasks than selecting participants randomly or through the reversed-auction-based approaches.

We present a naming, addressing, and profile server (NAPS) as a middleware to bridge different platforms in Internet-of-Things (IoT) sensory environments. Given massive amount of heterogeneous devices deployed across different platforms, NAPS serves as the key module at the back-end data center to aid the efficient upstream sensory data collection, content-based data filtering and matching, and downstream efficient control by applications. While previous research efforts only focus on a specific standard or protocol, we aim to design a middleware component servicing dynamic application needs, and sensors/actuators deployment and configurations across different platforms. Specifically, we propose a complete design of NAPS, including its key functionalities, system flows, interfaces, and individual module design. We further propose a unique device naming and addressing convention, and show its applicability to a few widely-used standards and protocols. We also propose an efficient identifier generation scheme; and we demonstrate a full implementation of the above designs with a case study, including a service registration portal. Finally, performance evaluation is done against the system throughput.

The ever increasing technological advances in embedded systems engineering, together with the proliferation of small-size sensor designs and deployment, have enabled smart devices to recognize daily human-based actions, activities, and interactions. Therefore, inferring a vast variety of user-device-based activities from diverse contexts obtained by a series of sensory observations has drawn much interest to the research area of ubiquitous sensing. The existence and awareness of context provides the capability of being conscious of physical environments or situations around users, and this allows network services to respond proactively and intelligently based on such awareness. Hence, with the evolution of smartphones, software developers have been empowered to create context-aware applications for recognizing human-centric or community based innovative social and cognitive activities in any situation and at any location. However, the middleware services provided in mobile devices have limited resources in terms of power, memory, and bandwidth compared to the capabilities of PCs and servers. Also, power is a major restriction on implementation of context-aware applications. Mobile device batteries do not last a long time while operating sensor(s) constantly. To this end, this article extensively surveys the emerging concepts of context awareness in mobile platforms by providing up-to-date literature and future research directives. We also point out the challenges faced in this regard and enlighten them by offering possible solutions.

The Internet of Things (IoT) is part of the Internet of the future and will comprise billions of intelligent communicating "things" or Internet Connected Objects (ICO) which will have sensing, actuating, and data processing capabilities. Each ICO will have one or more embedded sensors that will capture potentially enormous amounts of data. The sensors and related data streams can be clustered physically or virtually, which raises the challenge of searching and selecting the right sensors for a query in an efficient and effective way. This paper proposes a context-aware sensor search, selection and ranking model, called CASSARAM, to address the challenge of efficiently selecting a subset of relevant sensors out of a large set of sensors with similar functionality and capabilities. CASSARAM takes into account user preferences and considers a broad range of sensor characteristics, such as reliability, accuracy, location, battery life, and many more. The paper highlights the importance of sensor search, selection and ranking for the IoT, identifies important characteristics of both sensors and data capture processes, and discusses how semantic and quantitative reasoning can be combined together. This work also addresses challenges such as efficient distributed sensor search and relational-expression based filtering. CASSARAM testing and performance evaluation results are presented and discussed., IEEE sensors Journal, 2013. arXiv admin note: text overlap with arXiv:1303.2447

Crowd saliency prediction refers to predicting where people look at in crowd scene. Humans have remarkable ability to rapidly direct their gaze to select visual information of interest when looking at a visual scene. Until now, research efforts are still focused on that which type of feature is representative for crowd saliency, and which type of learning model is the robust one for crowd saliency prediction. In this paper, we propose a Random Forest (RF) based crowd saliency prediction approach with optimal feature combination, i.e., the Feature Combination Selection for Crowd Saliency (FCSCS) framework. More specifically, we first define two representative crowd saliency features: FaceSizeDiff and FacePoseDiff. Next, we adopt the Random Forest (RF) algorithm to construct our saliency learning model. Then, we evaluate the performance of crowd saliency prediction classifiers with different feature combinations (fifteen combinations in our experiments). Those selected features include low-level features (i.e., color, intensity, orientation), four existing crowd features (i.e., face size, face density, frontal face, profile face) and two new defined features (i.e., FaceSizeDiff and FacePoseDiff). Finally, we obtain the optimal feature combination that is most suitable for crowd saliency prediction. We conduct extensive experiments and empirical evaluation to demonstrate the satisfactory performance of our approach.

Participatory Crowdsourcing System (PCS) is now an important source of information for big data applications. In PCS, required environmental parameters are collected and shared by ordinary citizens. And the limited range of human activities and limited power or capability of smart devices make the PCS facing a lot of problems. But the wide use of robots, such as Unmanned Aerial Vehicle (UAV), can effectively alleviate the current problems in PCS. Lots of precision sensors can be installed into the robots, which makes sensory data collected by robots more trustful. However, the power and data transmission capability of robots are also limited. Meanwhile, it is much more expensive to use robots than recruit human participants. Thus, it will be most useful and necessary to use robots in PCS, let the robots and human participants working together, rather than using robots to replace the human participants. Obviously, the cooperation of robots and human participants will also make PCS facing new challenges, such as how to use the least amount of robots to reduce costs, etc.. In this paper, we analyze the different application scenarios of robots-aided PCS, and propose possible solutions to the existing problems. Finally, we discuss possible future work for robots-aided PCS.

A large amount of unstructured data are being continuously generated by mobile applications, e.g., Facebook, Twitter and Foursquare, due to the popularity of smart devices, e.g., iPad, iWatch and smartphone. These data are undoubtedly of great value but need to be first well stored and then analyzed. Nevertheless, traditional relational database management system (RDBMS) is not well qualified for this job. As a result, NoSQL databases are proposed and become very popular. Judging from the present situations, RDMBSs play a better role in managing relational data, and thus application developers have to face the situation of co-existence of both RDBMS and NoSQL databases, and knowing the details of the underlying data distributions and structures of different databases. To this end, in this paper, we aim to propose a unified, transparent query engine, as a middleware, called "ZQL", by using MySQL and Hive as underlying two exemplary databases, however its usage goes far beyond these two databases. ZQL aims to support application development by hiding the specific details of both NoSQL databases and RDBMS. Finally, extensive experimental results show the effectiveness and functionalities of the proposed and implemented ZQL middleware.

Cyber Physical Systems: Architectures, Protocols and Applications helps you understand the basic principles and key supporting standards of CPS. It analyzes different CPS applications from the bottom up, extracting the common characters that form a vertical structure. It presents mobile sensing platforms and their applications toward interrelated paradigms, highlighting and briefly discussing different types of mobile sensing platforms and the functionalities they offer. It then looks at the naming, addressing, and profile services of CPS and proposes a middleware component to meet the requirements of dynamic applications and sensors/actuators deployment/configurations across different platforms. The middle chapters of the book present a context-aware sensor search, selection, and ranking model which addresses the challenge of efficiently selecting a subset of relevant sensors out of a large set of sensors with similar functionality and capabilities. The authors consider various topics in the energy management of CPS and propose a novel energy-efficient framework. They also present the fundamental networking technologies of CPS and focus on machine-to-machine communications for CPS, specifically the open technologies such as IPv6-based solutions that can be integrated into IoT and enable wireless sensor communications. In the book's final chapters, the authors bring you up to date on mobile cloud computing (MCC) research activities that enhance the capabilities of resource-constrained smart devices in CPS sensory environments. They also present a few representative CPS applications, including connected healthcare, gaming in public transport crowds, and a series of MCC-enabled emerging CPS applications. You will find that these application fields fully demonstrate the great potential of applying CPS in public life.

Node density and connectivity of large-scale multi-channel cognitive radio networks (CRNs) are studied in this paper. Current existing researches on connectivity of CRNs are mainly conducted under the assumption of one licensed channel and ignore the spectrum sensing errors. Thus, in this paper, practical assumptions such as spatially varying spectrum holes for secondary users (SUs), multiple channels licensed for the primary network, correct detection and false alarm probabilities (Pd and Pf) are considered to derive the diluted density of SUs in closed-form for the secondary network. The sensing errors of SUs and amount of licensed channels of primary users (PUs) are found to have a significant impact on the density of active SUs. Increasing density of PUs considerably reduces the node density of active SUs. Sequentially connectivity of the secondary network is investigated by applying percolation theory and complex network. The critical condition for percolation of the secondary network coexisting with primary network is then estimated analytically. We study the impacts of node density, protection radius, amount of licensed channels of the primary network on connectivity of the secondary network. It is shown that more licensed channels can loosen the condition for SUs percolation. Also, sensing errors significantly affect the connectivity of SUs. However, we found that more accurate sensing carried by SUs (larger Pd and smaller Pf) does not necessarily bring better connectivity of the secondary network.

Participatory sensing campaigns encourage ordinary people to collect and share sensing data by using their portable smart devices like smartphones, iPad and iWatch. However, compared with static sensors that are deployed in the particular places by people that can provide high-quality data, the quality of sensing data that are collected from smart devices is uncontrollable, and it is costly to pay data usage without knowing the credibility of the collected data. In this paper, we propose to leverage the participant's reputation to maximize total data credibility under budget constraint for participatory sensing systems. Specifically, we formally define a maximum sensing data credibility problem and present a trustable participants selection algorithm to obtain a near optimal solution. Provided the real data set, it has been shown by simulation results that our proposed algorithm selects more credible sensing data than that of the other two comparison algorithms.

Crowdsourcing systems, by using smart devices like smartphones and iPad, have been widely used in various domains, but are currently facing new challenges. On one hand, different tasks offer different amount of incentive budgets in multitask systems, thus those tasks pay more should be satisfied preferentially. On the other hand, even if there are not enough budget for a sensing task, the system should also try to provide as much sensory data as possible to obtain potential clients. Furthermore, it is challenging to select an “optimal” set of participants as data contributors due to the above two points. In this paper, first, we introduce a metric to describe how the collected sensory data can quantify task's the multi-dimensional data requirements, in terms of data distribution in spatiotemporal domains. Second, we propose a “task priority model” based on incentive budget, to explicitly quantify the relationship between the incentive budget usage and task priority. Then, we present a quality of information (QoI) aware participant selection approach as a suboptimal solution to the defined optimization problem. Finally, we compare our proposed scheme with existing methods via extensive simulations based on the real movement traces of ordinary citizens in Beijing. Extensive simulation results well justify the effectiveness and robustness of our approach.

Detecting emergent events and monitoring civil infrastructures in modern metropolitan cities by participatory sensing have recently been identified as a critical part of the public service management. With fast information distribution, multimedia messages (e.g., sound, images, videos, and texts) collected from citizens' smart devices can provide useful information to infer such emergencies by processing application level queries initialized from end terminals. This requires to establish an efficient, real-time processing systems for participatory sensing that can cope with both the dynamic queries and a variety of information with diverse attributes. To this end, in this paper, we first design a distributed and real-time query framework for event-based stream processing in participatory sensing, including a Storm-based real-time query engine, a messaging queue on Kafka, and a data persistence module based on HBase. Second, a dynamic indexing division method that is aware of the change of query attributes and volume is proposed. Third, we implement an application for civil infrastructure monitoring, and finally we evaluate the performance of proposed framework compared with existing approaches, simulation results of which show its advantages.

This a commonly headache issue in our life, that is before dinners, meetings, movies, visits, or even only a walk, we all hesitate or finally annoyed on question "What am I gonna wear today?" Imaging such a scenario, a fashionable lady is called to meet her boyfriend when she is working in the office. Right after, she is going back home to well prepare for the date. After several tryons and comparisons, she picks a simple but elegant shirt. She is standing in front of a full-length mirror, and trying clothes one by one for hours. Meanwhile, her boyfriend is urging her to come over. But still, there are no trousers to match it. Finally, both of them may get a bad mood or even worth, they quarrel with each other for not coming on time. This is a common problem for almost everyone, especially for lady, that they have serious difficulties to make decision on which clothes to wear, or match their clothes, every time, every day. But now, a gospel, named the Smart Dressing Mirror, designed by the Lab of Intelligent Computing at Beijing Institute of Technology, China, appears for you, with some of fantastic functionalities that you need. This is also what we introduced today, a future ideas of your life in the new era of mart. Generally, this Smart Dressing Mirror is nearly the same as the traditional full-length mirror in every home in appearance, and the only difference is that the smart dressing mirror has some smart functions, which can interactive with you or be friend with you. Given that all your clothes are digitalized firstly after you bought it, and the digitalized image can be captured from the homepage of the clothes shop or pictured and normalized by the camera hidden in the top of the smart mirror. After that, with simple gestures, you can interactive with the smart mirror, and picking those digitalized clothes as you wish. You can see whether it fits or not. Furthermore, with its equipped machine learning and computer vision technologies, the Smart Mirror will recommend you a bunch of fashionable and suitable upper/lower clothes for you. For example: 1) when you choose an upper/lower clothes, the system will recommend you a bunch of matched lower/upper ones, 2) when you tell the mirror which scenario you will go for, it will also suggest you a series of clothes. At the same time, clothes from e-Shopping stores will also be recommended. You can buy it immediately, and it will be delivered to you as soon as possible. This is what we offer, a fantastic smart dressing mirror, to replace the traditional mirror in every home, in the near future.

Recommender systems have changed the way people originally find products, information, and even their social circles. However, most existing research activities neglect its time-varying feature, i.e., the growing input data, the change of user behaviors. In order to sustain the high accuracy of recommendations, systems have to be updated regularly. However, the more often the update proceeds, the more cost of time and other computational resources. Thus, it is critical to strike the balance between accuracy and cost. In this paper, we propose an adaptive recommender system by using feedback control frameworks. The proposed solution continuously monitors its changes and estimates the loss of performance (in terms of accuracy) from two perspectives: data problem(data aging and data deficient) in training set, and changes of user behavior by “revisiting ratio”. When the benefit of performing an update exceeds the cost of resources, the system update itself. Theoretical analysis and extensive results by using a real data set are supplemented to show the advantages of the proposed system.

Hadoop is emerging as one of the leading frameworks used by enterprises to help make better business decisions on large data sets. Virtualization technology brings plenty of benefits to Hadoop, including higher resource utilization and cluster reliability. However, these benefits mean nothing to users if unacceptable performance degradation happens from physical to virtual platform. Existing efforts on virtualized Hadoop performance find that improper configurations of network and storage with open sourced virtual deployment cause huge overhead on system performance. However, complexity of hardware and software including virtualization configurations and various scale of deployment also makes performance tuning still too hard a practice to execute. To span that gap of virtualized Hadoop adoption, in this paper, we propose a performance diagnostic methodology that integrates statistical analysis from different layers, and design a heuristic performance diagnostic tool which evaluates the validity and correctness of virtualized Hadoop by analyzing the job traces of popular big data benchmarks. By using this tool, users could quickly identify the bottleneck according to hints provided by this tool, further confirm the diagnosis by referring to performance utilities provided by guest OS and hypervisor, and continue tuning performance for virtualized Hadoop by multiple runs of this tool.

Smart devices are supporting emerging types of location-based gaming applications to attain collaborations among collocated users in public places like transports. However, they are facing many challenges like the timely performance of back-end game servers for runtime game operation, unreliable cellular network connections, and opportunistic and dynamic local environment. To address these limitations, in this paper, we propose “CrowdMoG”, a Crowd based Mobile Gaming platform, to identify and match the nearby passengers on the move according to their associated gaming preferences, and provides the smooth session handoff to enable the continuity of existing game plays when participants leave the game due to different mobility patterns. We therefore describe the potential, along with challenges and opportunities that open up the new dimension for the entire research community to redesign and examine a tradition problem, fundamentally transforming it into a new era of mobile gaming experience.

Although Wireless Sensor Networks (WSN) have emerged with a great potential to change our life, especially with ubiquitous sensing and sensory data access through Sensing-as-a-service ( $\text{S}^{2}$ aaS) models, there are still many fundamental technical challenges that need to be addressed to realize this potential. These challenges include the complexity and diversity of applications, the needs for highly flexible and secure configuration management mechanisms, the need to design programmable and novel sensory techniques, and the needs to balance local intelligence and scalable cloud-based intelligence to satisfy real-time service objectives.

In this paper, we propose an in-situ Crowd-aware Sensing Platform, called "CrowdMon", which envisions the cooperation among mobile users in highly crowded urban areas such as metro and square. CrowdMon establishes a spontaneous connection from co-located users in a semantic proximity and enables them to share contextual information such as location, ambient music, and mood of places. To the best of our knowledge, CrowdMon is the first attempt to support crowd-aware services at a platform level. We show interesting use cases of CrowdMon and an initial system design to realize the crowd-based context sharing.

In this paper, we present “HealthKiosk”, a family-based healthcare system, to provide rich contextual information and alerting mechanisms for continuous monitoring of health conditions. HealthKiosk significantly improves the reach and quality of regional and community healthcare services, and could potentially minimize the efforts of care professionals for the chronical illness, not only applicable for the elderly but also children and young fitness trainers. The proposed architecture leverages the RESTful design style, nevertheless introducing a novel development of the “sensor proxy”, both in a stationary PC style and as a mobile widget. The sensor proxy behaves not only as a bridge between the raw sensor readings and the kiosk controller, but also as a data processing logic to integrate, correlate, and transform a variety of data from multiple data sources to an integrated XML format. We demonstrate the system performance by showing two complete case studies for both stationary and mobile sensor proxies used in the developed HealthKiosk system. Finally, the proposed solution has been piloted and deployed at the Peking University People's Hospital (PKUPH) for diabetes patients, via building an evidence-based clinical care solution focusing on chronic disease management in China.