The effective segmentation of 3D data is crucial for a wide range of industrial applications, especially for detecting subtle defects in the field of integrated circuits (IC). Ceramic package substrates (CPS), as an important electronic material, are essential in IC packaging owing to their superior physical and chemical properties. However, the complex structure and minor defects of CPS, along with the absence of a publically available dataset, significantly hinder the development of CPS surface defect detection. In this study, we construct a high-quality point cloud dataset for 3D segmentation of surface defects in CPS, i.e., CPS3D-Seg, which has the best point resolution and precision compared to existing 3D industrial datasets. CPS3D-Seg consists of 1300 point cloud samples under 20 product categories, and each sample provides accurate point-level annotations. Meanwhile, we conduct a comprehensive benchmark based on SOTA point cloud segmentation algorithms to validate the effectiveness of CPS3D-Seg. Additionally, we propose a novel 3D segmentation method based on causal inference (CINet), which quantifies potential confounders in point clouds through Structural Refine (SR) and Quality Assessment (QA) Modules. Extensive experiments demonstrate that CINet significantly outperforms existing algorithms in both mIoU and accuracy.

Due to the widespread use of integrated circuits (IC) package substrates, especially ceramic package substrates (CPS), the industry has raised stringent quality evaluation standards. However, defective packaging substrate samples are scarce and there are many types of defects, so it is difficult for existing semantic segmentation methods to obtain accurate and generalized results on the CPS images. In order to solve the above problems, this article proposes an effective few-shot segmentation method, named dynamic cross characterization network (DCCNet), which can segment untrained CPS defect species using only a small number of labeled CPS samples. First, we introduce a cross sets attention mechanism to enhance the interconnection within the category and better distinguish the differences between the categories. Then, the introduction of the DC block dynamically represents the features, enhances the sensitivity of the features, and reduces the disturbance of differences between classes. Finally, we propose a TB block to deal with feature loss and interclass obfuscation during dimensional change. In addition, we propose a new CPS few-shot segmentation dataset CPSAD-FS to evaluate the proposed DCCNet. Through a large number of comparative experiments and ablation experiments, we have clearly evaluated the state-of-the-art performance of our DCCNet on the CPSAD-FS dataset and verified the effectiveness of each block.

The Segment Anything Model 2 (SAM2) has revolutionized prompt-driven visual segmentation, yet its integration with Neural Radiance Field (NeRF) for object segmentation remains constrained by static scene assumptions and Low Dynamic Range (LDR) input limitations. This paper introduces SA4D-HDR, a novel framework that lifts SAM2’s ability to 4D HDR scenes, enabling accurate 4D segmentation of arbitrary objects. Based on 4D HDR NeRF prior, we reconstruct temporally coherent HDR scenes from multi-exposure inputs. A sigmoid-based tone mapping function then projects the HDR scenes to LDR domain, enabling SAM2 to generate precise object masks within the reconstructed scenes. Subsequently, manual segmentation prompts in single-view input are fed into SAM2 to generate sequential 2D segmentation masks, which guide the learning of a mask deformation field network. This network models 4D geometric transformations in voxel space, while a lightweight decoder predicts masks at arbitrary timestamps. To enhance geometric consistency, we introduce an annealing jitter noise mechanism that injects controlled perturbations during training, mitigating alignment errors between NeRF-rendered geometries and SAM2-generated 2D masks. Our method establishes an effective segmentation framework for 4D HDR scenes with neural radiance fields, bridging the semantic gap between 2D visual foundation models and 4D HDR scenes while achieving state-of-the-art accuracy on HDR datasets.

Accurate detection of small objects is vital for analyzing traffic scenes in intelligent transportation systems. Advances in deep learning have greatly improved the detection of small objects in traffic environments. This study targets traffic scene extraction from images taken by unmanned aerial vehicles (UAVs) to enhance small object detection. We introduce FAS-Fusion, a selection and fusion network based on feature adaptation, which employs an optimized feature pyramid network (FPN) to refine feature extraction and capture critical details needed to identify small objects like vehicles from aerial views. This framework strengthens the model’s ability to effectively interpret feature data. Key components, including the Attention-Based Feature Adaptive Fusion (AFAFM) and Small Objects Attention (SOAT) modules, apply feature filtering to boost detection precision in aerial imagery. Furthermore, the Adaptive Receptive-Field Feature Enhancement Module (ARFEM) intelligently assesses feature relevance within the network’s receptive field to optimize feature representation. Extensive experiments on UAV datasets demonstrate that our approach substantially improves accuracy and precision in small object detection, offering significant benefits for traffic management and the development of smarter transportation systems.

As the core of the modern electronics industry, integrated circuits (IC) involve highly complex design and manufacturing processes, with the design and fabrication of the package substrates particularly impacting the circuit’s performance and reliability. Therefore, defect detection and classification of integrated circuits package substrates (ICPS) are crucial in IC production. Addressing issues such as the scarcity of data and the challenges in data perception for ICPS, we propose a Fine-grained Region Perception Network (FRPNet) to achieve multi-view perception and precise few-shot classification of ICPS. Specifically, FRPNet consists of three modules: the Category-Perceptive Interaction Module, responsible for feature aggregation perception during class simulation changes; the Fine-Grained Region Aggregation Module, which observes the regions of interest from multiple views and ensures intra-class connectivity; and the Localization Refinement Module, which enhances positional information to ensure the stability of features from local to global scales. Additionally, we construct a CPS2D-FSC dataset comprising single-layer and multi-layer ICPS. We conducted extensive experiments in CPS2D-FSC to validate FRPNet, including comparisons with SOTA algorithms and ablation studies, demonstrating the superiority of our algorithm and the effectiveness of each module.

Defect classification in metallic materials is critical for industrial quality assurance, yet existing few-shot learning (FSL) methods, primarily designed for natural images, struggle to address material heterogeneity, subtle defect variations, and the coexistence of surface/internal flaws in manufacturing scenarios. To bridge this gap, we propose MEDNet, a novel FSL framework integrating a Memory-Enhanced Irrelevance Elimination Module (MEIE) and a Dual-modality Feature Discrimination Module (DFD). The MEIE module employs a dynamically updated memory bank to isolate and suppress task-irrelevant features (e.g., material textures or background noise), thereby enhancing defect-specific discriminative representations. The DFD module leverages Maximum Mean Discrepancy (MMD) to align cross-modal distributions between support and query sets, ensuring robustness against material diversity and limited annotations.To support industrial defect analysis, we integrate a metal defect few-shot classification dataset (MD-FSC) comprising 27 defect categories, spanning surface cracks, internal pores and other defect types across diverse alloy types. Extensive experiments demonstrate MEMNet’s superiority: it achieves state-of-the-art accuracy of 77.29% (1-shot) and 83.13% (5-shot), outperforming existing FSL benchmarks by 6.21% and 4.92%. Ablation studies further validate the necessity of MEIE for feature purification and the dual-modality discrimination for material-agnostic generalization. Code and Dataset are publicly available at https://github.com/Zzhen266/MD-FSC

With the advent of network function virtualization (NFV), many users resort to network service provisioning through virtual network function instances (VNFIs) run on the standard physical server in clouds. Following this trend, NFV markets are emerging, which allow a user to procure VNFIs from cloud service providers (CSPs). In such procurement process, it is a significant challenge to ensure differential privacy and truthfulness while explicitly considering dynamic resource provisioning, location sensitiveness and budget of each VNFI. As such, we design a differentially private and truthful reverse auction with dynamic resource provisioning (PTRA-DRP) to resolve the VNFI procurement (VNFIP) problem. To allow dynamic resource provisioning, PTRA-DRP enables CSPs to submit a set of bids and accept as many as possible, and decides the provisioning VNFIs based on the auction outcomes. To be specific, we first devise a greedy heuristic approach to select the set of the winning bids in a differentially privacy-preserving manner. Next, we design a pricing strategy to compute the charges of CSPs, aiming to guarantee truthfulness. Strict theoretical analysis proves that PTRA-DRP can ensure differential privacy, truthfulness, individual rationality, computational efficiency and approximate social cost minimization. Extensive simulations also demonstrate the effectiveness and efficiency of PTRA-DRP.

Recommendation systems have primarily been limited to research on a single dataset compared to natural language processing and computer vision, which have seen tremendous growth in transferable tasks. Existing approaches for recommendation systems need to be more scalable to arbitrary tasks, given that previous research efforts on transferable recommendations have only yielded brief explorations and neglected systematic studies of sequential tasks. In this regard, we propose the interaction subgraph sequential topology-aware network (ISTN), which overcomes this limitation, enabling transferable sequence recommendations. ISTN performs subgraph sampling and node labeling of user interactions, captures the topological features of the user interaction sequences with the sequential topology auto-encoder, and employs the sequential preference decoupling module to decouple user interaction sequences for transferable adaptive granularity modeling of user preferences. ISTN requires no fine-tuning, and its knowledge transfer capability from the training dataset to the new dataset delivers accurate, individualized recommendation results. ISTN outperforms state-of-the-art performance in transferable contexts with only minor performance degradation compared to the traditional baseline, as shown in Yelp, MovieLens, and Foursquare experiments.

Mobile edge computation (MEC) has recently emerged as a promising computing paradigm for supporting latency-sensitive mobile applications. Due to the limited resources of the edge servers (ESs), efficient resource allocation mechanisms are key to realize the MEC paradigm. In such a resource allocation process, it is a significant challenge to guarantee truthfulness while enabling flexible task offloading and satisfying the locality constraint. To address such a challenge, we propose a truthful auction-based resource allocation mechanism with flexible task offloading (TARFO) in an MEC system. Specifically, we first design the minimum delay task graph partitioning algorithm, aiming at calculating the minimum completion time and the task offloading solutions under different resource profiles. Based on this algorithm, for each smart mobile device (SMD), we further determine the set of feasible non-dominated resource profiles and the corresponding task offloading solutions. We next propose an efficient primal-dual approximation winning bid selection algorithm to determine the set of the winning bids and a critical value based pricing algorithm to calculate the payments of the winning bids. Strict theoretical analysis demonstrates TARFO can ensure truthfulness, individual rationality, computational efficiency and a smaller approximation ratio. Simulation results verify the effectiveness and efficiency of TARFO.

Unmanned Aerial Vehicle (UAV)-enabled mobile edge computing has been proposed as an efficient task-offloading solution for user equipments (UEs). Nevertheless, the presence of heterogeneous UAVs makes centralized navigation policies impractical. Decentralized navigation policies also face significant challenges in knowledge sharing among heterogeneous UAVs. To address this, we present the soft hierarchical deep reinforcement learning network (SHDRLN) and dual-end federated reinforcement learning (DFRL) as a decentralized navigation policy solution. It enhances overall task-offloading energy efficiency for UAVs while facilitating knowledge sharing. Specifically, SHDRLN, a hierarchical DRL network based on maximum entropy learning, reduces policy differences among UAVs by abstracting atomic actions into generic skills. Simultaneously, it maximizes the average efficiency of all UAVs, optimizing coverage for UEs and minimizing task-offloading waiting time. DFRL, a federated learning (FL) algorithm, aggregates policy knowledge at the cloud server and filters it at the UAV end, enabling adaptive learning of navigation policy knowledge suitable for the UAV's performance parameters. Extensive simulations demonstrate that the proposed solution not only outperforms other baseline algorithms in overall energy efficiency but also achieves more stable navigation policy learning under different levels of heterogeneity of different UAV performance parameters.

Mobile crowdsensing leverages the power of a vast group of participants to collect sensory data, thus presenting an economical solution for data collection. However, due to the variability among participants, the quality of sensory data varies significantly, making it crucial to extract truthful information from sensory data of differing quality. Additionally, given the fixed time and monetary costs for the participants, they typically only perform a subset of tasks. As a result, the datasets collected in real-world scenarios are usually sparse. Current truth discovery methods struggle to adapt to datasets with varying sparsity, especially when dealing with sparse datasets. In this article, we propose an adaptive Hypergraph-based EM truth discovery method, HGEM. The HGEM algorithm leverages the topological characteristics of hypergraphs to model sparse datasets, thereby improving its performance in evaluating the reliability of participants and the true value of the event to be observed. Experiments based on simulated and real-world scenarios demonstrate that HGEM consistently achieves higher predictive accuracy.

As a core component of the Internet of Vehicles, reasoning about the trajectory of pedestrians or vehicles in complex road conditions plays a critical role in autonomous driving and socially aware robotic navigation. Most existing methods do not adequately consider the effects of heterogeneous traffic agents. Toward this end, we propose the traffic trajectory prediction algorithm based on the convolutional attention network (TraGCAN) to predict the trajectories of heterogeneous traffic agents in dense traffic. The algorithm of the proposed method examines the behavior of different traffic agents in terms of both time and space dimensions to identify their movement patterns and interactions. We construct the spatial relationship of traffic agents as a graph structure and introduce a graph convolutional network to extract spatial interactions. In addition, we design a spatial attention mechanism to adaptively calculate weights for all spatial interactions to capture different influences from neighboring agents. To improve the accuracy of trajectory prediction, the algorithm considers the influence of the heterogeneous characteristics of traffic agents on their motion behaviors. We evaluated the performance of the proposed TraGCAN on heterogeneous traffic data sets, and the results demonstrate that the error of TraGCAN is reduced by 15% compared to existing methods.

Automatic defect detection on the metal surface is a vital task for product inspection in industrial assembly lines or production processes. Owing to miscellaneous patterns of defects, interclass similarity, intraclass difference, and fewer defect samples, achieving accurate and automatic detection remains a big challenge. What is more, since the rising demand for production efficiency, real-time detection is increasingly desirable. This article proposes a semantic prior and extremely efficient dilated convolution network, named SPEED, for pixel-wise detection on the metal surface, which aims to address the aforementioned issues. The architecture of SPEED involves the following: 1) a semantic prior (SP) branch, with shallow layer and prior mapping module to capture low-level details; and 2) an extremely efficient dilation (EED) branch, with lightweight bottleneck to obtain high-level context. Furthermore, an aggregation module is designed to fuse both types of feature representation. Additionally, different level features of bottleneck are fused to improve the segmentation performance. Experimental results on three metal surface defect datasets indicate that the proposed method outperforms the state-of-the-art approaches in terms of the mean intersection of union, model parameters, FLOPs, and FPS. More specifically, SPEED achieves 92.34% mIoU on NEU-Seg, 88.65% mIoU on Severstal Strip Steel, and 63.91% mIoU on MT Defect.

With the increasing popularity of cab services such as Didi and Uber, cities are faced with the challenge of high carbon emissions and traffic congestion. Ride-sharing services, as a novel green mode of transportation, have emerged as a key technology in smart transportation for addressing these problems. The implementation of ride-sharing is predicated on an accurate ridership demand forecasting model, which can effectively prevent vehicle resource waste, alleviate traffic congestion, and reduce carbon emissions. In this paper, a periodic attentional graph convolutional spatio-temporal network model (PAG-TSN) is proposed to predict regional ridership demand. Specifically, the model is trained using a large amount of GPS data and user demand data collected by the travel service provider. PAG-TSN consists of two parts: the bicomponent attention graph convolution model (BAT-GCN) and the periodic attentional gated recurrent unit model (PA-GRU). The former uses GCN to extract spatial features from pointwise and edgewise graphs; the latter uses the spatial feature vectors extracted from the former with external information as input, and uses GRU to extract temporal features from feature data of different periods, and finally uses attention mechanism and POI requirement correlation to integrate the extracted spatio-temporal information to derive prediction results. Extensive experiments and evaluations on the CD2Date and XA2Date datasets show that PAG-TSN outperforms other baseline models in accurately predicting regional ridership demand, with MAPE and RMSE values of 0.1147 and 5.56, respectively.

Monocular 3D reconstruction is to reconstruct the shape of object and its other information from a single RGB image. In 3D reconstruction, polygon mesh, with detailed surface information and low computational cost, is the most prevalent expression form obtained from deep learning models. However, the state-of-the-art schemes fail to directly generate well-structured meshes, and we identify that most meshes have severe Vertices Clustering (VC) and Illegal Twist (IT) problems. By analyzing the mesh deformation process, we pinpoint that the inappropriate usage of Chamfer Distance (CD) loss is a root cause of VC and IT problems in deep learning model. In this article, we initially demonstrate these two problems induced by CD loss with visual examples and quantitative analyses. Then, we propose a fine-grained reconstruction method CD 2 by employing Chamfer distance twice to perform a plausible and adaptive deformation. Extensive experiments on two 3D datasets and comparisons with five latest schemes demonstrate that our CD 2 directly generates a well-structured mesh and outperforms others in terms of several quantitative metrics.

Deep learning-based recommendation systems have made significant strides in recent years. However, the problem of recommendation systems' generalizability has not been solved. After the training phase, most current models can only solve problems on a particular dataset and are not as generalizable as NLP and CV models. Therefore, a large amount of computing power is required to make conventional recommendation models available to different trades. In real-world scenarios, offline retailers often opt out of recommendation algorithms due to a lack of computer capacity, which puts them at a competitive disadvantage. As a result, we propose an Interaction Graph Auto-encoder Network (IGA) based on topology-aware to address the transferable recommendation problem. IGA is composed primarily of the following components: Interaction Feature Subgraph Extraction, Subgraph Node Labeling, Subgraph Interaction Auto-encoder, and Interaction Preference Attention Network. IGA can transfer knowledge from the training dataset to the new dataset without fine-tuning and give users reliable, personalized recommendation results. Experiments on the MovieLens, Douban, LastFM, and Book-Crossing datasets demonstrate that IGA outperforms state-of-the-art approaches in transferable scenarios. Additionally, IGA requires fewer computing power and is highly adaptable across datasets.

Predicting treatment medicines is a key task in many intelligent healthcare systems. Prediction of treatment medicines can assist doctors in making informed prescription decisions for patients according to their Electronic Health Records (EHRs). However, predicting treatment medicines is a challenging task due to the following reasons: (1) heterogeneous nature of EHR data that typically includes laboratory results, treatment records, disease conditions, and demographic information; (2) complex correlations among EHR sequences, including inter-correlations between sequences and temporal intra-correlations within each sequence; (3) temporal dynamics of these correlations changing with disease progression. In this paper, we predict treatment medicines for patients with dual adaptive sequential networks (DASNet). Specifically, DASNet is designed with three components. First, a decomposed adaptive long short-term memory network (DA-LSTM) is designed to capture the intra- and inter-correlations in multiple heterogeneous temporal sequences. Then, we develop an attentive meta learning network (AT-MetaNet) to learn dynamic weight parameters for DA-LSTM, thus enabling it to model various correlation structures. Finally, we employ an attentive fusion network (AT-FuNet) to incorporate historical information and collectively fuse representation embeddings of heterogeneous data to predict treatment medicines. Our results on the public MIMIC-III dataset covering 11 medical conditions demonstrate that the proposed end-to-end model can achieve the state-of-the-art prediction performance while providing clinically useful insights.

Numerous state-of-the-art recommendation frameworks employ deep neural networks in Collaborative Filtering (CF). In this paper, we propose a cross feature fusion neural network (CFFNN) for the enhancement of CF. Existing studies overlook either user preferences for various item features or the relationship between item features and user features. To solve this problem, we construct a cross feature fusion network to enable the fusion of user features and item features as well as a self-attention network to determine users’ preferences for items. Specifically, we design a feature extraction layer with multiple MLP (Multilayer Perceptrons) modules to extract both user features and item features. Then, we introduce a cross feature fusion mechanism for an accurate determination of the relationship between different user-item interactions. The features of users and items are crossly embedded and then fed into a prediction network. The attention mechanism enables the model to focus on more effective features. The effectiveness of CFFNN model is demonstrated through extensive experiments on four real-world datasets. The experimental results indicate that CFFNN significantly outperforms the existing state-of-the-art models, with a relative improvement of 3.0 to 12.1 percent on hit ratio (HR) and normalized discounted cumulative gain (NDCG) compared with the baselines.

The current Internet content delivery model assumes strict mapping between a resource and its descriptor, e.g., a JPEG file and its URL. Content Distribution Networks (CDNs) extend it by replicating the same resources across multiple locations, and introducing multiple descriptors. The goal of this work is to build Web-LEGO, an opt-in service, to speedup webpages at client side. Our rationale is to replace the slow original content with fast similar or equal content. Further, we perform a reality check of this idea both in term of the prevalence of CDN-less websites, availability of similar content, and user perception of similar webpages via millions of scale automated tests and thousands of real users. Then, we devise Web-LEGO, and address natural concerns on content inconsistency and copyright infringements. The final evaluation shows that Web-LEGO brings significant improvements both in term of reduced Page Load Time (PLT) and user-perceived PLT. Specifically, CDN-less websites provide more room for speedup than CDN-hosted ones, i.e., 7x more in the median case. Besides, Web-LEGO achieves high visual accuracy (94.2%) and high scores from a paid survey: 92% of the feedback collected from 1,000 people confirm Web-LEGO's accuracy as well as positive interest in the service.

Past years have witnessed the rapid increasing number of smart devices and objects deployed in the urban environment. Leveraging helpful data generated by hundreds of millions of smart objects, a large number of services in the Internet of Things (IoT) are devised and developed to improve our urban life quality. However, uploading the unprecedented volume of sensing data from IoT sensors to the cloud directly can lead to huge unnecessary consumption and hurt the quality of IoT services. This work leverages the fog architecture to devise a task-driven data offloading (TDO) algorithm in urban IoT services. Specifically, a three-layer urban IoT service architecture is proposed, and the TDO process is formulated as a combination optimization problem taking task deadlines and abilities of fog devices into consideration. Then, we prove the TDO problem is NP-hard, and the G-TDO algorithm is devised to solve it with a careful designed utility function. Also, we propose RG-TDO algorithm to improve the G-TDO algorithm considering the overlaps of tasks. Finally, we demonstrate the significant performance of the proposed algorithms with extensive evaluations based on real-world data set.

Point-of-Interest (POI) demand modeling in urban regions is critical for building smart cities with various applications, e.g., business location selection and urban planning. However, existing work does not fully utilize human mobility data and ignores the interactive-aware information. In this work, we design a refined POI demand modeling framework, named RePiDeM, to identify region POI demands based on multi-source data, including cellular data, POI data, satellite image, geographic data, etc. Specifically, we introduce a Cellular Data (CD) based visit inference algorithm to estimate the POI visit probability based on human mobility and POI data. Further, to address the data sparsity issue, we design a multi-source attention neural collaborative filtering (MANCF) model to output region POI demands considering various aspect attention. We conduct extensive experiments on real-world data collected in the Chinese city Shenyang, which show that RePiDeM is effective for modeling region POI demands.

Image-based virtual try-on systems with the goal of transferring a desired clothing item onto the corresponding region of a person have made great strides recently, but challenges remain in generating realistic looking images that preserve both body and clothing details. Here we present a new virtual try-on network, called VTNFP, to synthesize photo-realistic images given the images of a clothed person and a target clothing item. In order to better preserve clothing and body features, VTNFP follows a three-stage design strategy. First, it transforms the target clothing into a warped form compatible with the pose of the given person. Next, it predicts a body segmentation map of the person wearing the target clothing, delineating body parts as well as clothing regions. Finally, the warped clothing, body segmentation map and given person image are fused together for fine-scale image synthesis. A key innovation of VTNFP is the body segmentation map prediction module, which provides critical information to guide image synthesis in regions where body parts and clothing intersects, and is very beneficial for preventing blurry pictures and preserving clothing and body part details. Experiments on a fashion dataset demonstrate that VTNFP generates substantially better results than state-of-the-art methods.

With the emerging of sharing economy, taxi crowdsourcing delivery could be a feasible solution for logistics companies to deliver packages efficiently and securely with a lower cost in the urban area. In this paper, we propose LSTM2V, a novel mobility pattern-aware task recommendation algorithm for taxi crowdsourcing delivery leveraging the long short-term memory and Markov model. Taking the mobility pattern into consideration, LSTM2V leverages both deep learning and probabilistic model to recommend the most suitable tasks to taxis. It mainly consists of two components - the feature window based Long Short-Term Memory neural network (LSTM-w) and SpatioTemporal Markov (STM) model. The taxi mobility pattern is predicted by LSTM-w, and STM is utilized to predict locations which taxis can visit in the future. Extensive evaluations with real taxi trajectory dataset show LSTM2V can predict the mobility pattern precisely, improve the multi-location prediction accuracy, and recommend tasks efficiently.

Existing works on human activity recognition predominantly consider single-person scenarios, which deviates significantly from real world where multiple people exist simultaneously. In this work, we leverage transfer learning, a deep learning technique, to present a framework (TL-HAR) that accurately detects multiple human activities; exploiting CSI of WiFi extracted from 802.11n. Specifically, for the first time we employ packet-level classification and image transformation together with transfer learning to classify complex scenario of multiple human activities. We design an algorithm that extracts activity based CSI using the variance of MIMO subcarriers. Subsequently, TL-HAR transforms CSI to images to capture correlation among subcarriers and use a deep Convolutional Neural Network (d-CNN) to extract representative features for the classification. We further reduce training complexity through transfer learning, that infers knowledge from a pre-trained model. Experimental results confirm the significance of our approach. We show that using transfer learning TL-HAR improves recognition accuracy to 96.7% and 99.1 % for single and multiple MIMO links.

Mobile crowdsensing (MCS) is a new data collection paradigm profiting from the human-centric cyber social computing. However, due to the humans' uncontrollable mobility, it raises severe concerns of data redundancy and poor data quality. In this paper, we propose a novel data gathering architecture based on mobile edge computing (MEC), which distributes computing resources [edge nodes (ENs)] in the sensing scenarios close to the mobile users, and thus enables significant improvements to handle users' frequent location changes and reduce the specified quantity of sensing tasks. Based on the MEC-enhanced architecture, we design a quality-aware sparse data collection (QSDC) algorithm in the MCS systems. In the ENs' part, QSDC exploits the implicit correlation (IC) among the sensing data to reduce the data redundancy and selects the appropriate users' group to ensure the spatiotemporal coverage of sensing grids. In the cloud server part, QSDC leverages the compressive sensing to recover the sensing data in the whole sensing area with high data quality. Extensive experiments verify the performance of QSDC based on real data sets under different experiment settings and demonstrate the effectiveness and availability of QSDC.

The ubiquity of ever-more-capable mobile devices, especially smartphones, brings forth participatory sensing to collect and interpret information. It can achieve unprecedented quantity of data. However, it is arduous to guarantee quality of data because everyone can contribute data without scrutinization. It is an important issue in quality-oriented participatory sensing. Our idea to address this issue is motivating participants to contribute accurate data for improving data quality directly. In this article, we propose a reputation-based incentive mechanism, RIM, to realize the idea. More specifically, we identify the participants who collect the accurate data and regard them as the reputable ones. Then, the reputable participants are granted a higher chance to obtain rewards so that other people will try to follow such users and become reputable as well. Namely, RIM can encourage and steer users to collect accurate data in the long term. We analyze our incentive mechanism by formalization and premise implications. For a feasibility study of participatory sensing and verification of the implications, we implement and deploy a participatory sensing application focusing on monitoring environmental noise in a specific location as a case study and conduct a simulation based on the case study to further evaluate the proposed incentive mechanism. The results from the case study and the simulation present that RIM can remarkably increase the quality of collected data in participatory sensing while corroborating our theoretical implications.

Privacy is a pivotal issue of mobile apps because there is a plethora of personal and sensitive information in smartphones. Many mechanisms and tools are proposed to detect and mitigate privacy leaks. However, they rarely consider users' preferences and expectations. Users hold various expectation towards different mobile apps. For example, users may allow a social app to access their photos rather than a game app because it goes beyond users' expectation to access personal photos. Therefore, we believe it is practical and beneficial to understand users' privacy expectations on various mobile apps and help them mitigate privacy risks introduced by smartphones. To achieve this objective, we propose and implement PriWe, a system based on crowdsourcing driven by users who contribute privacy permission settings of the apps installed on their smartphones. PriWe leverages the crowdsourced permission settings to understand users' privacy expectations and provides app specific recommendations to mitigate information leakage. We deployed PriWe in the real world for evaluation. According to the feedback of 78 users who evaluated our system and 422 participants who completed our survey, PriWe is able to make proper recommendations which can match participants' privacy expectations and are mostly accepted by users, thereby help them to mitigate privacy disclosure in smartphones.

Urban crowdsourced transportation, which can solve traffic problem within city, is a new scenario where citizens share vehicles to take passengers and packages while driving. Differing from the traditional location based crowdsourcing system (e.g., crowdsensing system), the task has to be completed with visiting two different locations (i.e., start and end points), so task allocation algorithms in crowdsensing cannot be leveraged in urban crowdsourced transportation directly. To solve this problem, we first prove that maximizing the crowdsourcing system’s profit (i.e., maximizing the total saved distance) is an NP-hard problem. We propose a heuristic greedy algorithm called Saving Most First (SMF) which is simple and effective in assigning tasks. Then, an optimized SMF based genetic algorithm (SMF-GA) is devised to jump out of the local optimal result. Finally, we demonstrate the performance of SMF and SMF-GA with extensive evaluations, based on a large scale real vehicle traces. The evaluation with large scale real dataset indicates that both SMF and SMF-GA algorithms outperform other benchmark algorithms in terms of saved distance, participant profits, etc.

Channel State Information of WiFi signal has attracted tremendous interests in recent years for activity identification. Although existing work can achieve desirable performance using different algorithms, similar system modules are often shared. In this paper, we first summarize and compare various techniques employed in different modules such as preprocessing, activity extraction, feature dimension reduction, and classification. Specifically, different feature reduction methods are applied in order to address the challenge of classifying various length signals and extracting representative abstractions, including manually selecting features and Dynamic Time Warping based classification with Principal Component Analysis. By targeting at multiple human activities, we then compare the performance of two common system structures from difference aspects. Experimental results show that it can be subjective and environment dependent by manually selecting particular features, while DTW based classification can be time consuming especially with larger dataset. In order to address these challenges, we propose a novel framework based on Deep Learning Network. Long Short Term Memory model, a type of Recurrent Neutral Network, is employed for time-series sequence classification. Extensive results show that it can achieve higher efficiency and accuracy.

with the strong impact of OTT (Over The Top) business in the mobile Internet era, operators urgently need to discover the user value information from massive data to help them provide personalized accurate services and expand business customer services. The construction of social user groups based on mobile communication data can help operators to accurately analyze customer social structures, thus promoting quality service and improving marketing quality. In this paper, we design a set of social group construction algorithm based on user behavior characteristics excavated from massive user data in mobile communication network. Due to the huge volume of mobile communication data sets, a parallel design based on MapReduce is exploited. The experimental results show that the ADBLINKw algorithm performs well on the efficiency and community detection quality.

Geomagnetic field signal has gained increasing wide investigated for indoor positioning problems. Because of the variation of magnetic signals and the sensor observation drift, the recent positioning technology development based on magnetic field or pedestrian dead reckoning (PDR) has been restricted. In addition, the accumulative error and cold-start problem can also cause huge positioning error. In this paper, we present a novel indoor localization approach, GeoLoc, for combining magnetic fingerprint matching and PDR by Kalman Filter. First, magnetic field intensity of every positions is gathered and a fingerprint map is built for matching. A candidate set of positions is introduced to include uncertainty and increase robustness for our estimation. With the squeezing of candidate sets, uncertainties of orientation and position estimation have been eliminated. Realistic experiment results show that GeoLoc successfully addresses accumulative error and cold-start problems by sensor data fusion. GeoLoc achieves a good estimation for both short (less than 17.5m) and long walking distance, and it can work in both offline and online real-time positioning. GeoLoc is able to achieve an online positioning accuracy of less than 1.2m, and an offline positioning accuracy of 0.3m only with a smart phone. GeoLoc only uses the built-in sensors of mobile phones, thus users can get their position only by using their phone.

An extensive set of research efforts have explored Channel State Information for human activity detection. By extracting CSI from a sequence of packets, one can statistically analyze the temporal variations embedded therein and recognize corresponding human activities. In this paper, we present Wi-Chase, a sensorless system based on CSI from ubiquitous WiFi packets for human activity detection. Different from existing schemes utilizing only CSI of one or a small subset of subcarriers, Wi-Chase fully utilizes all available subcarriers of the WiFi signal and incorporates variations in both their phases and magnitudes. As each subcarrier carries integral information that will improve the recognition accuracy because of detailed correlated information content in different subcarriers, we can achieve much higher detection accuracy. To the best of our knowledge, this is the first system that gathers information from all the subcarriers to identify and classify multiple activities. Our experimental results show that Wi-Chase is robust and achieves an average classification accuracy greater than 97% for multiple communication links.

Patent litigation not only covers legal and technical issues, it is also a key consideration for managers of high-technology (high-tech) companies when making strategic decisions. Patent litigation influences the market value of high-tech companies. However, this raises unique challenges. To this end, in this paper, we develop a novel recommendation framework to solve the problem of litigation risk prediction. We will introduce a specific type of patent-related litigation, that is, Section 337 investigations, which prohibit all acts of unfair competition, or any unfair trade practices, when exporting products to the United States. To build this recommendation framework, we collect and exploit a large amount of published information related to almost all Section 337 investigation cases. This study has two aims: (1) to predict the litigation risk in a specific industry category for high-tech companies and (2) to predict the litigation risk from competitors for high-tech companies. These aims can be achieved by mining historical investigation cases and related patents. Specifically, we propose two methods to meet the needs of both aims: a proximal slope one predictor and a time-aware predictor. Several factors are considered in the proposed methods, including the litigation risk if a company wants to enter a new market and the risk that a potential competitor would file a lawsuit against the new entrant. Comparative experiments using real-world data demonstrate that the proposed methods outperform several baselines with a significant margin.

With the ever expanding mobile device ecosystem, mobile users face a vast and constantly growing application pool. At the same time, in our daily life, waiting occurs regularly at different places such as shopping centers, where mobile applications become the de facto means to consume the time periods. In this paper, we propose a novel application recommendation system that utilizes human activity information at different places, to better match the applications with the characteristics of the users current contexts. Specifically, we design a place/application matching model and present two application list recommending algorithms with bounded approximation ratio. We also implement the recommendation system on real mobile phones and conduct field studies to show its feasibility. Our experimental and simulation results show that the proposed schemes can achieve satisfactory results.

Vehicular ad hoc networks, as a special case of delay tolerant networks, have become increasingly attractive to academia and industry. Different from most of the work in this field, which has focused on short periods of transient opportunistic contacts, in our previous work, we have analyzed the position data of a large set of urban private vehicles in Changsha, China and proposed a Location based Urban Vehicular network (LUV) utilizing the stable connections among vehicles. Place serves as a central message exchange and routing component in LUV that is critical in providing relatively reliable network connections. In this paper, we present a simple threshold based approach for identifying the places or vehicle aggregation areas, in an urban environment. We perform experimental study over a real set of data gathered over three months for 8900 vehicles and show the method is effective.

Opportunistic networks are lack of end-to-end paths between source nodes and destination nodes, so the communications are mainly carried out by the “store-carry-forward” strategy. Selfish behaviors of rejecting packet relay requests will severely worsen the network performance. Incentive is an efficient way to reduce selfish behaviors, and hence improves the reliability and robustness of the networks. In this paper, we propose the Node-dependence-based Dynamic gaming Incentive (NDI) algorithm, which exploits the dynamic repeated gaming to motivate nodes relaying packets for other nodes. The NDI algorithm presents a mechanism of tolerating selfish behaviors of nodes. Reward and punishment methods are also designed based on the node dependence degree. Simulation results show that the NDI algorithm is effective on increase the delivery ratio and decrease average latency when there are a lot of selfish nodes in the opportunistic networks.

Wireless sensor networks are usually densely deployed, and it is quite common for sensors to gather and transmit redundant information, which results in unnecessary energy consumption. Sleep scheduling is quite helpful for reducing overall energy consumption of the network, and thus prolongs the network lifetime. In this paper, we propose an Information Entropy Approach for Sleep Scheduling (IEASS). Information entropy is exploited in the algorithm to characterize the correlation of data which is used for determining the eligibility of node sleeping. The main objective of IEASS is to achieve adaptive coverage while keeping network connectivity. From the simulation results, IEASS performs well on coverage ratio and coverage degree with much less active sensor nodes. Moreover, IEASS achieves high flexibility by adjusting the algorithm parameters.

Sparse sensor networks have emerged in recent studies. Relaying data with the help of mobile elements seems an effective way to bridge the gaps in such networks. In this paper, we propose the Grid-Based Mobile Element Scheduling (GBMES) approach that schedules a mobile element (ME) to periodically g