Silent speech interface (SSI) enables users to interact with their devices without making audible sounds, thus preventing potential eavesdropping or disruptions to others. Recent advancements in acoustic sensing technology have made SSI on mobile devices highly promising, requiring no hardware modifications and operating in a non-contact manner. However, one major challenge faced by existing acoustic sensing-based SSI is its limited sensing range, typically less than 7cm. Users often need to speak in close proximity to the speakers/microphones, severely constraining its applicability on mobile devices. In this paper, we introduce EchoLip, which can significantly increase the sensing range and enhance long-term usability in real-world settings. EchoLip utilizes the smartphone’s two built-in speakers and microphones to transmit and receive mutually orthogonal wave signals to capture multi-view information. Then, a specially designed signal processing pipeline and neural network are used to extract fine-grained features that adapt to different angles and distances. We also design a lip movement monitoring algorithm to handle various interference. We evaluate EchoLip on 20 individuals using a set of 500 sentences. EchoLip achieves an average Word Error Rate of 13.9% and 19.7% at 15cm and 40cm. Evaluations in various scenarios further validate the robustness of EchoLip.

Accurately recognizing facial expressions and emotions at any time and in any place can significantly improve people's quality of life and mental well-being. However, existing methods lack the convenient capability for long-term monitoring in the wild environment. In this paper, we introduce EchoExpress, an in-the-wild emotion-related facial expression recognition system that works in an unobtrusive, low-power, and privacy-friendly way. EchoExpress uses two speakers and two microphones mounted on a glass-frame for transmitting and receiving mutually orthogonal wave signals. Concurrently, a unique attention mechanism dynamically extracts crucial features, enabling the capture of nuanced facial expressions and emotions. Furthermore, we introduce an open-set filtering mechanism with a specially designed loss function, which effectively filters out irrelevant actions, thereby reducing the risk of misidentification. Finally, a semi-supervised training method is employed to address the significant variability in wild expressions across different individuals. In extensive testing, EchoExpress achieves an accuracy of 84% in a laboratory environment and over 75% in real-world conditions. We believe that EchoExpress can serve as an unobtrusive and reliable way to monitor facial expressions.

Sequential recommendation aims to predict users’ next interactions by analyzing their historical behaviour sequences. One common approach is modeling both long-term and short-term interests together, but there are still two major challenges that need to be addressed. The first challenge is that repeated learning of neighbor nodes in GNN will blur the long-term interests of users. The second challenge is that existing models often fuse a user’s diverse interests together to predict their behaviour. However, each interest may independently affect the user’s next interaction. In this work, we propose modeling long-term interests by Directed Acyclic Graph Neural Network(LDAG) to address the above challenges. Specifically, we use metric learning to transform the item sequence into a compact item-item directed acyclic graph. We also propose a directed acyclic graph convolution and a readout based on the sink set in DAG to aggregate and extract users’ interests. Finally, users’ long-term interests and diverse short-term interests are combined for the prediction of users’ next behaviour. We conduct extensive experiments on three real-world datasets. Experimental results demonstrate that our proposed model outperforms current state-of-the-art methods. Further experiments illustrate the rationality and effectiveness of our proposed method.

Wireless-based gesture recognition provides an effective input method for exergames. However, previous works in wireless-based gesture recognition systems mainly recognize one primary user's gestures. In the multi-player scenario, the mutual interference between users makes it difficult to predict multiple players' gestures individually. To address this challenge, we propose a flexible FMCW-radar-based system, RFDual, which enables real-time cross-domain gesture sequence recognition for two players. To eliminate the mutual interference between users, we extract a new feature type, biased range-velocity spectrum (BRVS), which only depends on a target user. We then propose customized preprocessing methods (cropping and stationary component removal) to produce environment-independent and position-independent inputs. To enhance RFDual's resistance to unseen users and articulating speeds, we design effective data augmentation methods, sequence concatenating, and randomizing. RFDual is evaluated with a dataset containing only unseen gesture sequences and achieves a gesture error rate of 1.41%. Extensive experimental results show the impressive robustness of RFDual for data in new domains, including new users, articulating speeds, positions, and environments. These results demonstrate the great potential of RFDual in practical applications like two-player exergames and gesture/activity recognition for drivers and passengers in the cab.

With deep learning, millimeter-wave radar-based gesture recognition applications have achieved satisfactory results. However, most existing approaches highly rely on highquality labeled data, and they suffer from severe over-fitting when labeled data are scarce. To end this, we present RadarAE, a novel representation learning framework for radar sensing applications. RadarAE learns sophisticated representations from massive low-cost unlabeled radar data, which enables accurate gesture recognition with few labeled data. To achieve this goal, we first meticulously observe the characteristics of raw radar data and extract an effective feature, Spatio-Temporal Motion Map (STMM). Then we borrow the key principle of Masked Autoencoders (MAE), a self-supervised learning technique for images, and propose an MAE-like model to learn useful representations from STMM. To adapt RadarAE to radar sensing applications, we present a series of customization techniques, including data augmentation, optimized model structure, and adaptive pretraining method. With the learned high-level representations, gesture recognition models can achieve superior performance in few-shot scenarios. Experiment results show that our model can achieve 79.1%, 92.1%, 97.8%, and 99.5% recognition accuracy in the 1, 2, 4, and 8-shot scenarios, respectively, where x-shot refers to the number of labeled samples for each gesture type. The source codes and dataset are made publicly available11https://githuh.com/Ela-Boska/RadarAE.

Variational AutoEncoder (VAE) is a popular deep generative framework with a solid theoretical basis. There are many research efforts on improving VAE. Among the existing works, a recently proposed deterministic Regularized AutoEncoder (RAE) provides a new scheme for generative modeling. RAE fixes the variance of the inferred Gaussian approximate posterior distribution as a hyperparameter, and substitutes the stochastic encoder by injecting noise into the input of a deterministic decoder. However, the deterministic RAE has three limitations: 1) RAE needs to fit the variance; 2) RAE requires ex-post density estimation to ensure sample quality; 3) RAE employs an additional gradient regularization to ensure training smoothness. Thus, it raises an interesting research question: Can we maintain the flexibility of variational inference while simplifying VAE, and at the same time ensuring a smooth training process to obtain good generative performance? Based on the above motivation, in this paper, we propose a novel Semi-deterministic and Contrastive Variational Graph autoencoder (SCVG) for item recommendation. The core design of SCVG is to learn the variance of the approximate Gaussian posterior distribution in a semi-deterministic manner by aggregating inferred mean vectors from other connected nodes via graph convolution operation. We analyze the expressive power of SCVG for the Weisfeiler-Lehman graph isomorphism test, and we deduce the simplified form of the evidence lower bound of SCVG. Besides, we introduce an efficient contrastive regularization instead of gradient regularization. We empirically show that the contrastive regularization makes learned user/item latent representation more personalized and helps to smooth the training process. We conduct extensive experiments on three real-world datasets to show the superiority of our model over state-of-the-art methods for the item recommendation task. Codes are available at https://github.com/syxkason/SCVG.

In collaborative filtering, it is an important way to make full use of social information to improve the recommendation quality, which has been proved to be effective because user behavior will be affected by her friends. However, existing works leverage the social relationship to aggregate user features from friends' historical behavior sequences in a user-levelindirect paradigm. A significant defect of the indirect paradigm is that it ignores the temporal relationships between behavior events across users. In this paper, we propose a novel time-aware sequential recommendation framework called Social Temporal Excitation Networks (STEN), which introduces temporal point processes to model the fine-grained impact of friends' behaviors on the user's dynamic interests in an event-leveldirect paradigm. Moreover, we propose to decompose the temporal effect in sequential recommendation into social mutual temporal effect and ego temporal effect. Specifically, we employ a social heterogeneous graph embedding layer to refine user representation via structural information. To enhance temporal information propagation, STEN directly extracts the fine-grained temporal mutual influence of friends' behaviors through themutually exciting temporal network. Besides, user's dynamic interests are captured through theself-exciting temporal network. Extensive experiments on three real-world datasets show that STEN outperforms state-of-the-art baseline methods. Moreover, STEN provides event-level recommendation explainability, which is also illustrated experimentally.

In modern society, wireless signals are ubiquitous in various indoor environments, such as living houses, offices and shop malls, facilitating human living in various aspects. Action recognition is a technique in roaring demand in the field of human-computer interaction. Whereas previous research studies propose various methods to action recognition using wireless signals, action recognition in locations with limited data is still very challengeable. To realize decent action recognition with the help of wireless signals, we propose WiLAR, a location-adapted action recognition system based on WiFi, which enables action detection, segmentation and recognition with commodity WiFi devices in locations with different amounts of data. WiLAR extracts informative features from fine-grained WiFi channel state information (CSI), and then feeds features into elaborately designed deep learning models to realize action recognition in different locations. In our dedicated experiments, WiLAR achieves average 97% accuracy workout recognition in locations with plenty of data, and also outperforms other recognition models in locations with limited training data.

Representation learning is the keystone for collaborative filtering. The learned representations should reflect both explicit factors that are revealed by extrinsic attributes such as movies' genres, books' authors, and implicit factors that are implicated in the collaborative signal. Existing methods fail to decompose these two types of factors, making it difficult to infer the deep motivations behind user behaviors, and thus suffer from sub-optimal solutions. In this paper, we propose Decomposed Collaborative Filtering (DCF) to address the above problems. For the explicit representation learning, we devise a user-specific relation aggregator to aggregate the most important attributes. For the implicit part, we propose Decomposed Graph Convolutional Network (DGCN), which decomposes users and items into multiple factor-level representations, then utilizes factor-level attention and attentive relation aggregation to model implicit factors behind collaborative signals in fine-grained level. Moreover, to reflect more diverse implicit factors, we augment the model with disagreement regularization. We conduct experiments on three public accessible datasets and the results demonstrate the significant improvement of our method over several state-of-the-art baselines. Further studies verify the efficacy and interpretability benefits bought from the fine-grained implicit relation modeling. Our Code is available on https://github.com/cmaxhao/DCF.

Sequential recommendation as an emerging topic has attracted increasing attention due to its important practical significance. Models based on deep learning and attention mechanism have achieved good performance in sequential recommendation. Recently, the generative models based on Variational Autoencoder (VAE) have shown the unique advantage in collaborative filtering. In particular, the sequential VAE model as a recurrent version of VAE can effectively capture temporal dependencies among items in user sequence and perform sequential recommendation. However, VAE-based models suffer from a common limitation that the representational ability of the obtained approximate posterior distribution is limited, resulting in lower quality of generated samples. This is especially true for generating sequences. To solve the above problem, in this work, we propose a novel method called Adversarial and Contrastive Variational Autoencoder (ACVAE) for sequential recommendation. Specifically, we first introduce the adversarial training for sequence generation under the Adversarial Variational Bayes (AVB) framework, which enables our model to generate high-quality latent variables. Then, we employ the contrastive loss. The latent variables will be able to learn more personalized and salient characteristics by minimizing the contrastive loss. Besides, when encoding the sequence, we apply a recurrent and convolutional structure to capture global and local relationships in the sequence. Finally, we conduct extensive experiments on four real-world datasets. The experimental results show that our proposed ACVAE model outperforms other state-of-the-art methods.

BACKGROUND: Accurately forecasting the prognosis could improve cervical cancer management, however, the currently used clinical features are difficult to provide enough information. The aim of this study is to improve forecasting capability by developing a miRNAs-based machine learning survival prediction model. RESULTS: The expression characteristics of miRNAs were chosen as features for model development. The cervical cancer miRNA expression data was obtained from The Cancer Genome Atlas database. Preprocessing, including unquantified data removal, missing value imputation, samples normalization, log transformation, and feature scaling, was performed. In total, 42 survival-related miRNAs were identified by Cox Proportional-Hazards analysis. The patients were optimally clustered into four groups with three different 5-years survival outcome (≥ 90%, ≈ 65%, ≤ 40%) by K-means clustering algorithm base on top 10 survival-related miRNAs. According to the K-means clustering result, a prediction model with high performance was established. The pathways analysis indicated that the miRNAs used play roles involved in the regulation of cancer stem cells. CONCLUSION: A miRNAs-based machine learning cervical cancer survival prediction model was developed that robustly stratifies cervical cancer patients into high survival rate (5-years survival rate ≥ 90%), moderate survival rate (5-years survival rate ≈ 65%), and low survival rate (5-years survival rate ≤ 40%).

Handwritten signature verification techniques, which can facilitate user authentication and enable secure information exchange, are still important in property safety. However, on-line automatic handwritten signature verification usually requires dynamic handwritten patterns captured by a special device, such as a sensor-instrumented pen, a tablet or a smartwatch on the dominant hand. This paper presents SonarSign, an on-line handwritten signature verification system based on inaudible acoustic signals. The key insight is to use acoustic signals to capture the dynamic handwritten signature patterns for verification. Particularly, SonarSign exploits the built-in speakers and microphones of smartphones to transmit a specially designed training sequence and record the corresponding echo for channel impulse response (CIR) estimation, respectively. Based on the sensitivity of CIR to the tiny surrounding environment changes including handwritten signature actions, SonarSign designs an attentional multi-modal Siamese network for end-to-end signatures verification. First, multi-modal CIR streams are fused to extract representative signature pattern features from spatio-temporal dimensions. Then an attentional Siamese network is elaborated to verify whether the given two signatures are from the same signatory. Extensive experiments in real-world scenarios show that SonarSign can achieve accurate and robust signatures verification with an AUC (Area Under ROC (Receiver Operating Characteristic) Curve) of 98.02% and an EER (Equal Error Rate) of 5.79% for unseen users.

Tag-aware recommender systems (TRS) utilize rich tagging records to better depict user portraits and item features. Recently, many efforts have been done to improve TRS with neural networks. However, these solutions rustically rely on the tag-based features for recommendation, which is insufficient to ease the sparsity, ambiguity and redundancy issues introduced by tags, thus hindering the recommendation performance. In this paper, we propose a novel tag-aware recommendation model named Tag Graph Convolutional Network (TGCN), which leverages the contextual semantics of multi-hop neighbors in the user-tag-item graph to alleviate the above issues. Specifically, TGCN first employs type-aware neighbor sampling and aggregation operation to learn the type-specific neighborhood representations. Then we leverage attention mechanism to discriminate the importance of different node types and creatively employ Convolutional Neural Network (CNN) as type-level aggregator to perform vertical and horizontal convolutions for modeling multi-granular feature interactions. Besides, a TransTag regularization function is proposed to accurately identify user's substantive preference. Extensive experiments on three public datasets and a real industrial dataset show that TGCN significantly outperforms state-of-the-art baselines for tag-aware top-N recommendation.

Gesture recognition plays a fundamental role in emerging Human-Computer Interaction (HCI) paradigms. Recent advances in wireless sensing show promise for device-free and pervasive gesture recognition. Among them, RFID has gained much attention given its low-cost, light-weight and pervasiveness, but pioneer studies on RFID sensing still suffer two major problems when it comes to gesture recognition. The first is they are only evaluated on simple whole-body activities, rather than complex and fine-grained hand gestures. The second is they can not effectively work without retraining in new domains, i.e. new users or environments. To tackle these problems, in this paper, we propose RFree-GR, a domain-independent RFID system for complex and fine-grained gesture recognition. First of all, we exploit signals from the multi-tag array to profile the sophisticated spatio-temporal changes of hand gestures. Then, we elaborate a Multimodal Convolutional Neural Network (MCNN) to aggregate information across signals and abstract complex spatio-temporal patterns. Furthermore, we introduce an adversarial model to our deep learning architecture to remove domain-specific information while retaining information relevant to gesture recognition. We extensively evaluate RFree-GR on 16 commonly used American Sign Language (ASL) words. The average accuracy for new users and environments (new setup and new position) are $89.03%$, $90.21%$ and $88.38%$, respectively, significantly outperforming existing RFID based solutions, which demonstrates the superior effectiveness and generalizability of RFree-GR.

Factorization Machine (FM) is an effective solution for context-aware recommender systems (CARS) which models second-order feature interactions by inner product. However, it is insufficient to capture high-order and nonlinear interaction signals. While several recent efforts have enhanced FM with neural networks, they assume the embedding dimensions are independent from each other and model high-order interactions in a rather implicit manner. In this paper, we propose Convolutional Factorization Machine (CFM) to address above limitations. Specifically, CFM models second-order interactions with outer product, resulting in ''images'' which capture correlations between embedding dimensions. Then all generated ''images'' are stacked, forming an interaction cube. 3D convolution is applied above it to learn high-order interaction signals in an explicit approach. Besides, we also leverage a self-attention mechanism to perform the pooling of features to reduce time complexity. We conduct extensive experiments on three real-world datasets, demonstrating significant improvement of CFM over competing methods for context-aware top-k recommendation.

Smart retail stores open new possibilities for enabling a variety of physical analytics, such as users' shopping trajectories and preferences for certain items. This paper aims to excavate three kinds of relations in physical stores, i.e. user-item, user-user and item-item, which provide abundant information for enhancing users' shopping experiences and boosting retailers' sales. We present ShopEye, a hybrid RFID and smartwatch system to delve into these relations in an implicit and non-intrusive manner. The intuition is that inertial sensors embedded in smartwatches and RFID tags attached to items can capture the user behaviors and the item motions, respectively. ShopEye first pairs users with corresponding items according to correlations between inertial signals and RFID signals, and then incorporates these pairs with the motion behaviors of users to further profile user-user and item-item relations. We have tested the system extensively in our lab environment which mimics the real retail store. Experimental results demonstrate the effectiveness and robustness of ShopEye in excavating these relations.

Automatic sign language recognition is an important milestone in facilitating the communication between the deaf community and hearing people. Existing approaches are either intrusive or susceptible to ambient environments and user diversity. Moreover, most of them perform only isolated word recognition, not sentence-level sequence translation. In this paper, we present MyoSign, a deep learning based system that enables end-to-end American Sign Language (ASL) recognition at both word and sentence levels. We leverage a lightweight wearable device which can provide inertial and electromyography signals to non-intrusively capture signs. First, we propose a multimodal Convolutional Neural Network (CNN) to abstract representations from inputs of different sensory modalities. Then, a bidirectional Long Short Term Memory (LSTM) is exploited to model temporal dependences. On the top of the networks, we employ Connectionist Temporal Classification (CTC) to get around temporal segments and achieve end-to-end continuous sign language recognition. We evaluate MyoSign on 70 commonly used ASL words and 100 ASL sentences from 15 volunteers. Our system achieves an average accuracy of 93.7% at word-level and 93.1% at sentence-level in user-independent settings. In addition, MyoSign can recognize sentences unseen in the training set with 92.4% accuracy. The encouraging results indicate that MyoSign can be a meaningful buildup in the advancement of sign language recognition.

Regular exercise offers numerous health benefits and contributes to a healthy lifestyle. Doing exercise at home is an attractive choice for many people due to its convenience and low cost. Motivated by this, we propose FitAssist in this paper, a household virtual fitness assistant capable of performing fine-grained exercise recognition and exercise quality assessment based on commercial WiFi devices. Unlike wearable devices based systems, this system is more comfortable and device-free. In addition, compared to previous Wi-Fi based exercise monitoring systems, whose performance attenuates seriously when users stand out of the First Fresnel Zone (FFZ), FitAssist does not require users to stand on or near the line of sight (LoS) path. To achieve this, FitAssist extracts features from the fine-grained WiFi channel state information (CSI) and enables both exercise recognition and user identification via deep learning techniques. Moreover, FitAssist can provide personalized workout assessment to help users obtain effective workout and prevent injury. Extensive experimental results in real settings show that FitAssist achieves average accuracies of 97% and 98% for exercise recognition and user identification respectively, as well as giving accurate and useful feedback in various scenarios, which proves its effectiveness and robustness.

Monitoring aperiodic activity sequence contributes a lot to home exercise guidance and sports experience but existing approaches are designed for quasi-period activity or isolated activity monitoring. There is a compelling need for contactless real-time auxiliary exercise system, especially for aperiodic activity sequence. In this paper, we present Wiga, a WiFi-based real-time contactless activity sequence recognition system, which can recognize activity sequences even for users who have not participated in the training phase. Wiga takes the fine-grained Channel State Information (CSI) as input and elaborates a deep learning network to map the motion-induced signal variations with the activity sequence. First, Wiga removes noise and redundancy of the raw CSI measurements. Then, after abstracting deep features with a Convolutional Neural Network (CNN), Wiga exploits a Long Short Term Memory (LSTM) network to model temporal dependencies of the sequence. In addition, Wiga employs the beam search method to get around error-prone temporal segments and obtains real-time activity sequence recognition. We evaluate Wiga with 17 yoga activities from 7 volunteers, and extensive experimental results show that Wiga achieves an average accuracy of 97.7% and 85.6% for trained and untrained users respectively with a recognition delay no more than 0.5s.

Smart wearable devices have become pervasive as they are portable and intelligent. The popular method to interact with it is touch-screen, which is error-prone and cumbersome due to its limited size. There are a few innovative works designing a virtual dial plate on the hand back, which need special-purpose sensors or microphones which may suffer from privacy leak. We propose MType, a system only employs sensors already built in the commercial-off-the-shelf (COTS) device with a magnetic ring to expand the interaction space between users and wearable devices. The core idea is to leverage the gravity sensor, linear accelerometer and magnetometer embedded in the standard smartwatch to detect gestures, capture input events and locate keystrokes on the opisthenar and palm. Besides, MType designs a runtime adaptation mechanism to handle the cold start problem and adapt to the variations over the time of usage. We implement MType on the COTS smartwatch and our extensive experiments in different scenarios show that the average accuracy of keystroke localization can reach 93% with a small size initial training set (3 samples for each key) at a low sampling rate (51Hz). Furthermore, when turning on the runtime adaptation mechanism and enlarging the training set, the accuracy can achieve 98%.

In recent years, location-based services have been widely applied not only in daily life but also in automation industries. As one of main location sensing technologies, RFID based localization has attracted increasing attention. Existing synthetic aperture RFID localization systems use the inverse correlation filter to reconstruct holograms and achieve satisfactory accuracy. However, these methods require accurate aperture positions for theoretical signal construction, while the ubiquitous aperture uncertainty in practice causes non-negligible performance degradation. In this paper, we present PEC, an accurate synthetic aperture RFID localization system with aperture position error compensation, which has a major advantage over the classic systems for no need to know the exact trajectory of the synthetic aperture. We first build a mathematical model for localization and merge all coherent received signals to estimate the tag position. Then we propose an iterative algorithm which can alternately estimate both the tag position and the aperture position error. We have implemented and evaluated PEC using commercial-off-the-shelf (COTS) RFID devices. Extensive experimental results show that it achieves the cm-level accuracy with aperture position error in noisy environments, which proves its effectiveness and robustness.

In recent years, radio frequency identification (RFID)-based approaches have been demonstrated to be a promising indoor localization techniques for many valuable applications, such as tracking tagged objects on the manufacturing lines, locating items in smart warehouses, and so on. In the near future, many applications will gain great benefits from knowing the positions of RFID-tagged objects. However, existing localization approaches often suffer from severe accuracy degradation in real-world environments due to the prevalent environmental interferences, such as the multipath effects. To this end, we designed an RFID-based localization system FaHo, which leverages a deep learning enhanced holographic technique for locating RFID tags accurately even in complex indoor environments. By carefully analyzing the features of the traditional holographic method, we created a new hologram-based algorithm called joint hologram, which yields a robust likelihood for each assumed position to be the true tag position. FaHo then adopts a deep convolutional neural network for analyzing the whole hologram, and subsequently estimate the true location of the RFID tag rather than simply seek for the largest-likelihood location. Furthermore, we implemented FaHo and evaluated its performance in several multipath-rich scenarios. The experimental results show that FaHo can achieve centimeter-level accuracy in both the lateral and radial directions using only one moving antenna. More importantly, our work also demonstrates that hologram-based localization is a highly effective technique for RFID indoor localization tasks.

At present, wireless sensing based gesture recognition is becoming a rising star due to its convenience and non-invasiveness without privacy issues, while the strict requirement of the deployment and surrounding environment is still an unavoidable issue which limits its development and generalization. Although there are some works involving the environmental variance, the changes of relative positions between devices and users are ignored. As one of the most popular wireless sensing methods, RFID is widely used in activity recognition with its stable low-level physical characters such as phase and RSS. Besides, the signals reflected from RFID tags intuitively delineate its movements. On the other hand, many interactive gesture-driven applications, such as gesture input for video games, have a paramount and unavoidable issue about the latency between completion of a gesture and its recognition. Inspired by deep learning, this paper presents a real-time ongoing gesture recognition system EUIGR, which efficiently integrates phase and RSS data streams, and extracts both environment and user invariant features. The proposed system seamlessly integrates CNNs (Convolutional Neural Networks) and LSTM (Long Short-Term Memory) to fuse RFID low-level physical characters and extract space-temporal information. Furthermore, with adversarial learning, EUIGR suppresses environment-related factors and the user-specific features, and obtains strong robustness to individual diversity and decreases the environmental dependence. We also implement the system with COTS RFID devices, and extensive experimental results show the effectiveness and accuracy of EUIGR.

Online bookstores are capable of capturing readers preferences by analyzing click logs and transaction records, while physical bookstores or libraries still lack effective methods to gather reader behavioral data. Fortunately, the widespread use of mobile wearable devices and RFID technology opens up new possibilities for uncovering in-store experience. In this paper, we propose ReaderTrack, a system that integrates smartwatch and RFID to excavate interactions between readers and books. We first leverage inertial sensors of smartwatch and backscatter signals of RFID tags to infer reader behaviors and book motions, respectively. Then we associate readers with their corresponding books according to previously inferred behaviors and motions. We implement ReaderTrack with COTS devices and evaluate it extensively in our lab environment which mimics a typical reading room. Experimental results show the effectiveness and robustness of ReaderTrack in reader-book interaction reasoning.

In the future, libraries and warehouses will gain benefits from the spatial location of books and merchandises attached with RFID tags. Existing localization algorithms, however, usually focus on improving positioning accuracy or the ordering one for RFID tags on the same layer. Nevertheless, books or merchandises are placed on the multilayer in reality and the layer of RFID tagged object is also an important position indication. To this end, we design PRMS, an RFID based localization system which utilizes both phase and RSSI values of the backscattered signal provided by a single antenna to estimate the spatial position for RFID tags. Our basic idea is to gain initial estimated locations of RFID tags through a basic model which extracts the phase differences between received signals to locate tags. Then an advanced model is proposed to improve the positioning accuracy combined with RF hologram based on basic model. We further change traditional deployment of a single antenna to distinguish the features of RFID tags on multilayer and adopt a machine learning algorithm to get the layer information of tagged objects. The experiment results show that the average accuracy of layer detection and sorting at low tag spacing ($2\sim8$cm) are about 93% and 84% respectively.

2018 IEEE International Conference on Pervasive Computing and Communications (PerCom), Athens, Greece, March 19-23, 2018

Monitoring respiration and heartbeat contributes to disease prediction, sub-health diagnosis, exercise and sleep quality analysis, fatigue warning, and even emotion estimation. There is a compelling need for contactless, easy-to-deploy and long-term respiration and heartbeat monitoring. In this paper, we present CRH, an RFID-based contactless respiration and heartbeat monitoring system. The key insight is that the RFID signal fluctuation induced by the chest motion is synchronous with respiration and heartbeat. Therefore, CRH collects the temporal phase information from the tag array near or on body to extract respiration and heartbeat signals using a sequence of signal processing techniques. We propose a signal separation method based on multi-tag empirical mode decomposition (EMD) to obtain respiration rate and heart rate after preprocessing. Furthermore, CRH can also detect intense motions and abnormal respiration. We implement and evaluate CRH using Commercial Off-The-Shelf (COTS) RFID devices. Extensive experimental results in different scenarios show that CRH can achieve high accuracy for monitoring multi-user respiration and heart rates, validating its wide applicability and high reliability for contactless fine-grained respiration and heartbeat monitoring.

Gesture recognition is an innovative technology which is fundamentally reshaping the way people live, entertain and work. However, most gesture recognition systems focus on the recognition of simple gestures and ignore the full potential of sequential gestures involving a series of temporally-related simple actions in order. This paper presents SGRS, a battery-free, scalable and non-specific sequential gesture recognition system based on COTS RFID. The key insight is that finegrained phase information extracted from RF signals is capable of perceiving various gestures. In SGRS, we meticulously devise gesture recognition mechanism by incorporating the k-means based vector quantizer and string matching algorithm to enable precise and real-time sequential gesture identification. Moreover, an improved edit distance algorithm is proposed for suppressing individual diversity. We implement SGRS and comprehensively evaluate the performance by recognizing traffic command gestures of Chinese traffic police. Experimental result shows that SGRS achieves an average recognition accuracy of 96.2% with eight sequential gestures and is highly robust to both individual diversity and multipath effect.

Collaborative Filtering (CF), a well-known approach in producing recommender systems, has achieved wide use and excellent performance not only in research but also in industry. However, problems related to cold start and data sparsity have caused CF to attract an increasing amount of attention in efforts to solve these problems. Traditional approaches adopt side information to extract effective latent factors but still have some room for growth. Due to the strong characteristic of feature extraction in deep learning, many researchers have employed it with CF to extract effective representations and to enhance its performance in rating prediction. Based on this previous work, we propose a probabilistic model that combines a stacked denoising autoencoder and a convolutional neural network together with auxiliary side information (i.e, both from users and items) to extract users and items’ latent factors, respectively. Extensive experiments for four datasets demonstrate that our proposed model outperforms other traditional approaches and deep learning models making it state of the art.

Live migration of virtual network functions (VNF) is a powerful technique with benefits of server maintenance, resource management and dynamic workload re-balance, among others. Downtime and total migration time are mainly two vital indicators to describe the performance of the VNF live migration (VLM). Modern research has effectively reduced the downtime to zero for some specific VNFs (eg. virtual router). However, for general VNFs predominantly leveraging pre-copy approach, such as firewalls, network address translators (NAT), load balancers, etc., there still remain some intractable problems: inevitable service downtime and long migration time on account of large amount of data transferred during migration, both of which result in a severe performance degradation of VNF services. To resolve these issues, we present a solution called PV-QAT to accelerate the migration process for these general VNFs. The PV-QAT creatively exploits the Para-Virtualization (PV) driver to filter out the useless memory pages in the process of migration, and unprecedentedly applies QuickAssist Technology (QAT) to provide fast compression of memory pages with low overhead, The experimental results show that PV-QAT can significantly reduce 77.5% of downtime and 80.5% of total migration time on average when compared with original pre-copy migration of KVM.

Social recommendation which incorporates social information has attracted wide attention across both academia and industry for its superior performance. However, most existing approaches interpret social information in a heuristic manner which is not effective to capture the strong interplay between social connections and behaviors of users. This paper proposes SoGeM (SOcial based GEnerative Model) which simulates user's behaviors in a generative way and models intrinsic preferences and social influences of users simultaneously. Different from the most approaches that preassign similarity weights between friends, SoGeM learns the social influences automatically and quantitatively. Thus, the learnt influence has a probabilistic interpretation, for it is produced along with the generative process. We use Gibbs Sampling to train SoGeM and conduct comprehensive experiments on three real datasets. The results show that SoGeM outperforms other state-of-the-art approaches.

Innovative Human Machine Interface technologies are fundamentally reshaping the way people live, entertain and work. Passive RFID tags, benefiting from its wireless, inexpensive and battery-free sensing ability, are gradually being applied in new-style interaction interfaces, ranging from virtual touch screen to 3D mouse. This paper presents TagController, a universal wireless and battery-free remote controller with two types of interactive actions. The key insight is that the fine-grained phase information extracted from RF signals is capable of perceiving various actions. TagController can recognize 10 actions without any training or prestored profiles by executing a sequence of functional components, i.e. preprocessor, action detector and action recognizer. We have implemented TagController with COTS RFID devices and conducted substantial experiments in different scenarios. The results demonstrate that TagController can achieve an average recognition accuracy of 95.8% and 94.3% in the scenarios of one and two remote controllers, respectively, which promises its feasibility and robustness.

Workflow recognition is a key technique in the field of activity recognition with benefits of monitoring the step being performed in the workflow, detecting the missing step, and providing assistance to the performer of the workflow, among others. In this paper, we present an unobtrusive workflow recognition system called RFlow-ID, which is the first device-free, battery-free and privacy-preserving workflow recognition system based on RFID technique. RFlow-ID perceives the use and movement of associated objects in the workflow using fine-grained phase information extracted from low-level RF signal, and infers the most likely sequence of workflow activities via a VQ-HMM model. We implement RFlow-ID on COTS RFID devices and evaluate it through a common biomedical experiment. The results validate the high recognition accuracy and robustness of our system.

Through the investigation of the mathematical model of hologram-based indoor localization system using RFID, this paper reveals two potential deficiencies about accuracy and gives the machine learning interpretation of the model. Exploiting the methods from machine learning and the thought of hierarchy, the output accuracy and the efficiency of the model can be further boosted. Simulation and experiment show that the enhanced model can halve mean error and attain 9x execution speed improvement.

Internet of Things (IoT) is rather prevalent in many manufacturing and smart city applications, while localization is a premise for many other processes, varying from ordering objects in manufacturing lines to locating books on bookshelves. Radio Frequency Identification (RFID) based localization is of great interest in many IoT applications. Synthetic aperture RFID, due to its anti-noise capability and robustness against multipath distortion, is becoming a rising star in the field of localization. Existing systems achieve finer lateral resolution, whereas their radial accuracy is limited by the narrow bandwidth of RFID signal. In this paper, we present a novel synthetic aperture RFID localization method which combines RFID phase based ranging with synthetic aperture technology, to achieve a higher radial accuracy than the existing systems. With only one reader antenna and one 1-dimensional (1D) trajectory, a synthetic array is constructed to get an accurate localization result both in lateral and radial direction. Its core idea is to make full use of the coherence of all multi-frequency phase data and merge them into a unique ranging based likelihood function. To improve the accuracy, the relative phase is leveraged to eliminate phase offsets caused by the reader antenna, and the phase deviation from the angle-of-arrival response is calibrated by pre-processing. Then a weighted enhancement is fully exploited to further improve the localization performance. We evaluate its performance with commercial-off-the-shelf (COTS) RFID devices and the results show that it achieves median accuracy of 3cm in both lateral and radial direction. This novel promising method is suitable for locating tags placed densely in many IoT applications, such as test tubes in hospitals.

Summary Shop recommendation in large shopping malls is useful in the mobile internet era. With the maturity of indoor positioning technology, customers' indoor trajectories can be captured by radio frequency identification devices readers, which provides a new way to analyze customers' potential preferences. In this paper, we design three methods for the top‐N shop recommendation problem. The first method is an improved matrix factorization method fusing estimated prior customer preference matrix that is constructed by Session‐based Temporal Graph computing. The second method is a Bayesian personalized ranking method based on the first method. The third method is by tensor decomposition combined with Session‐based Temporal Graph. Besides, we exploit customer history radio frequency identification devices trajectory information to find customers' frequent paths and revise predicted rating values to improve recommendation accuracy. Our methods are effective in modeling customers' temporal dynamics. At the same time, our approach considers repeated recommendation of the same shop by designing rating update rules. The test dataset is formed byJoyCitycustomer behavior records.JoyCityis a large‐scale modern shopping center in downtown Shanghai, China. The results show that our approaches are effective and outperform previous state‐of‐the‐art approaches. Copyright © 2016 John Wiley & Sons, Ltd.

Radio Frequency Identification (RFID) is a rapidly developing digital identification technology that employs radio to collect identification information from RFID tags. In a typical RFID identification scenario, an RFID reader sends a request to RFID tags, and the RFID tags reply with the information pre-stored in their storages. In recent decades, many applications such as supply chain management, auto-ticking, human and animal tracking, smart hospital, etc. employ more and more RFID readers.

In this paper, we propose two novel approaches for recommendation in large shopping mall scenario. For matrix factorization approach, we construct a bias matrix utilizing graph computing which fuses user’s long-term and short-term preferences. We exploit user trajectories to mine user’s frequent paths and adopt to revamping rules to update ratings from the result of matrix factorization, thus solving the problem of re-predicting customer’s preference to all shops in a new time window. For tensor decomposition approach, we add time dimension and construct a customer-shop-time three dimensional tensor, predict ratings are from the slice of the approximate tensor. We evaluate the result by top N recall and precision rate. Our data set is made on JoyCity which is a real shopping mall in Shanghai, the result is encouraging and it shows that our approach is applicative.

In recent decades, a growing number of large-scale RFID systems are used in various applications. In such a system, it is not uncommon that multiple concurrent radio communications among RFID readers and tags cause serious inference (called collision in the RFID field). One important kind of method to achieve collision-free communication is to activate RFID readers in different time slots. Existing activation approaches for solving this problem are mainly centralized, which is impractical due to the lack of central server, one-point failure risk, and performance bottleneck. Some distributed algorithms are proposed recently, but failed to consider the adaptiveness of the identification, where all of the RFID readers need to participate in the coordination control even if they do not have communication requirements any more. As a result, the optimal identification performance cannot be achieved. In this paper, we propose an adaptive distributed reader activation approach called ADRA for large-scale RFID systems. We build a fine-grained conflict graph for different kinds of collisions. And then a shared permission based distributed approach is adopted to eliminate those collisions. We guarantee that the RFID readers that do not need to communicate any more are suspended and excluded from the execution of coordination eventually. Extensive simulation results show that our approach outperforms existing approaches in terms of execution time and message overhead.