• Frequency-aware temporal context guides adaptive spatial topology learning • Time2Vec encoding captures both periodic and non-periodic motion patterns • Achieves SOTA: 93.8 Kinetics-Skeleton • Ablation studies show 3.5 guidance • Efficient architecture with favorable parameter-accuracy trade-off Skeleton-based human activity recognition (HAR) has made significant progress through graph convolutional networks (GCNs) and Transformer architectures for spatiotemporal modeling. However, existing methods either employ predefined static graph topologies that cannot adapt to heterogeneous skeleton data or learn dynamic topologies based solely on local spatiotemporal features, thereby overlooking the global temporal frequency features of joint movements that are important for discovering semantically meaningful spatial relationships. We propose Frequency-Aware Topology Learning Graph Convolutional Network (FATL-GCN), a novel architecture that integrates frequency-aware temporal context to guide adaptive learning of spatial topology. Our approach leverages Time-to-Vector linear frequency encoding to capture both periodic and non-periodic motion patterns, employs frequency-guided topology learning to generate action-specific graphs through temporal-context-driven attention, and incorporates hierarchical multi-scale fusion for robust feature extraction across scales. Extensive experiments achieved top-1 accuracies of 93.8% (cross-subject) and 97.5% (cross-view) on NTU-60, 91.9% (cross-subject) and 93.1% (cross-setup) on NTU-120, and 51.7% on Kinetics-Skeleton. Ablation studies confirm the critical role of our components, with removing the dynamic graph topology causing a 3.5% accuracy drop and removing frequency-aware encoding causing a 2.1% drop.

Existing unsupervised domain adaption approaches for cross-user Wearable Human Activity Recognition (WHAR) typically assume that users utilize the uni-modal sensor deployment configuration and cannot transfer across different sensor modalities. In this paper, we consider the more realistic cross-modal wearable human activity recognition setting to investigate the unsupervised domain adaptation task. This new context presents two formidable challenges: (1) how to alleviate modality heterogeneity across users, and (2) how to explore cross-modal domain correlation for better unsupervised domain adaptation. We propose a cross-modal unsupervised domain Adaptation model with Class-Aware Sample Weight Learning (CASWL-Adapt) to address both challenges. First, a spherical modality discriminator is designed to capture modal-specific discriminative features of each user during domain adaptation, thus achieving a reduction of sample variance caused by modal heterogeneity. Given a user-specific modal, modality-independent domain-invariant features can be efficiently generated by the well-developed modality discrimination loss and adversarial training. Second, a class-aware weight network is devised to calculate sample weights through classification loss and activity class similarity for each sample. Furthermore, the network leverage end-to-end learning and meta-optimization update rules to explore inter-domain correlations. Cross-modal activity classes are expected to adaptively implement different weighting schemes based on their intrinsic bias characteristics to select the most appropriate samples for domain knowledge transfer. We demonstrate that CASWL-Adapt achieves state-of-the-art results on three challenging benchmarks: Epic-Kitchens, Multimodal-EA and RealWorld, especially effective for new users of unseen modality.

Depression has long been described as a common mental health disorder and a disease with a set of diagnostic criteria that influences the affected individuals' feelings and behavior. The prevalence of Internet use has augmented people’s openness to share their experiences and struggles, including mental health disorders on social media thus researchers have tried developing classification models for depression detection using various machine learning and deep learning techniques. In this research, we propose a deep learning architecture with an attention mechanism on CNN-BiLSTM (CBA) and provide a comparative analysis to benchmark well-known deep learning models using the public dataset namely CLEF2017. We found that along with F1 score, precision and recall it is also vital to consider the Area under the curve - Receiver operating characteristic curve (AUC-ROC) and Mathews Correlation Coefficient (MCC) metrics for evaluating depression classification models since the MCC considers all the four values of a confusion matrix. Based on our experiments, the CBA model outperforms the existing state of the art model with an overall accuracy of 96.71% and scores of 0.85 and 0.77 for AUC-ROC and MCC, respectively.

An accurate heart rate (HR) estimation using a radar sensor is challenging in real-life situations due to frequent changes in a person’s sitting position and posture. Furthermore, radar signal is affected by the presence of clutter noise, the harmonics associated with respiration (RR), and the movements of the body. To address these challenges, this article proposes a real-time beam steering algorithm and a signal processing technique based on Resonance Sparse Spectrum Decomposition (RSSD). Our beam steering method dynamically calculates the range–angle values of the target during the scanning phase and determines the target’s position at the beginning of each measurement cycle. This allows the beam-steered signal to be directed toward the individual, thereby improving the signal-to-noise ratio (SNR). We present a novel signal processing method based on RSSD that leverages sub-band energy distribution to optimize the quality factor (Q) and the subsequent extraction of HR using harmonics. The Q factor defines the resonance property of an oscillatory signal, and hence, signal components with similar center frequency bands but with a different quality factor, Q, can be separated and sparsely represented. The RSSD-based algorithm mitigates the effects of clutter and random body motion from the phase signal, which significantly enhances HR estimation accuracy. Comprehensive experiments performed under various realistic conditions demonstrate that the HR accuracy remains consistently high at 98.72% within a 4 m range across all azimuth angles.

The noncontact measurement and monitoring of human vital signs has evolved into a valuable tool for efficient health management. Because of the greater penetration capability through material and clothes, which is less affected by environmental conditions such as illumination, temperature, and humidity, mmWave radar has been extensively researched for human vital sign measurement in the past years. However, interference due to unwanted clutter, random body movement, and respiration harmonics make accurate retrieval of the heart rate (HR) difficult. This article proposes a resonance sparse spectrum decomposition (RSSD) algorithm and harmonics used algorithm (HUA) for accurate HR extraction. RSSD addresses the clutter and random body movement effects from phase signals, while HUA uses harmonics to extract HR accurately. A set of controlled experiments was conducted under different scenarios, and the proposed method is validated against ground truth HR/RR data collected by a smart vest. Our results show an accuracy of up to 98%–100% for distances up to 2 m. The method substantially improves HR estimation accuracy by effectively mitigating the effects of noise in the phase signal, even under heavy clutter and moderate body movement. Our results demonstrate that the proposed method effectively counters harmonic interference for accurate estimation of HR comparable to RR estimation up to a distance of 4 m from the radar sensor.

Deep Neural Networks (DNN) based collaborative filtering has been successful in recommending services by effectively generalizing graph-structured data. However, most existing approaches focus on first-order interactions. Although recent approaches have utilized high-order connectivity, they still limit themselves to simple interactions and ignore the pattern of structural sub-graphs/motifs. In this study, we first explore the commonly used motifs in the Mashup-API interaction bipartite graph and propose a dedicated algorithm to generate the motif adjacency matrix. We then propose a Motif-based Graph Attention Network for service recommendation (MGSR) that utilizes a motif-based attention mechanism to capture the high-order information of various motifs, and a Collaborative Filtering model to generate the recommendation prediction. We have conducted extensive experiments on ProgrammableWeb dataset and our results demonstrate the superior performance of our proposed framework over some state-of-the-art approaches.

Despite the continual increase in the number of Web-APIs available on the internet, it is still challenging for API consumers to discover appropriate Web-APIs that could satisfy requirements. One of the main reasons for this is that Web-APIs registered on online directories such as ProgrammableWeb are in general isolated, as they are registered by diverse providers independently and progressively, ignoring continuous interactions among these APIs, which could enhance their discoverability. In this paper, we propose a method for analyzing the Web service ecosystem, and a complex network-based approach for constructing evolving networks for Web-APIs that are capable of enhancing their discoverability. We first conduct a two-phase analysis: We investigate mashups and Web-APIs interactions in the service ecosystem, and analyze their popularity distributions, and quantitatively measure two key node attachment dimensions within the ecosystem: Preferential Attachment and Similarity. Based on the analysis, we propose two methods for constructing evolving Web-API networks using the theoretical procedures of theBarab asi-Albert and the Popularity-SimilarityOptimizationnetwork models. Finally, we comprehensively evaluate the networks and map their properties with service discoverability using the ProgrammableWeb datasets. The results presented in this work will serve as a practical guide for designing a complex network-based solution for Web-API discovery.

Many real-world networks including the World Wide Web and the Internet of Things are graphs in their abstract forms. Graph neural networks (GNNs) have emerged as the main solution for deep learning on graphs. Recently, tremendous effort has been made to enhance the performance and expressivity of GNNs. In this paper, we review the state-of-the-art graph neural network models and frameworks with a focus on the latest developments in graph representation learning. We propose a new taxonomy which divides general GNNs into recurrent GNNs, spectral GNNs, spatial GNNs and topology-aware GNNs. We will also discuss the inductive biases behind different categories of GNNs.

In recent years, various approaches for multiresident human activity recognition (HAR) in a smart indoor environment have been developed and improved along with the rapid development of sensors and AI technologies. Research in data stream-based online learning (OL) for multiresident HAR is relatively new and a majority of the existing works have been developed based on training batches of data that cannot recognize real-time activities. To address the challenges of OL for multiresident HAR, we propose a novel OL architecture based on a locally weighted ensemble detection-based adaptive random forest (LED-ARF) classifier. We conduct a comprehensive performance comparison of eight famous OL classification techniques and our LED-ARF method. The comparison is evaluated based on the two benchmarking CASAS and ARAS data sets. Our experimental results show that LED-ARF achieves the best performance with the highest robustness for online multiresident HAR.

Several techniques for human activity recognition (HAR) in a smart indoor environment have been developed and improved along with the rapid advancement of sensor technologies. However, recognizing multiple people’s activities is still challenging due to the complexity of their activities, such as parallel and collaborative activities. To address these challenges, we propose a transformer with a bidirectional gated recurrent unit (GRU) deep learning (DL) method, called TRANS-BiGRU, to efficiently learn and recognize different types of activities performed by multiple residents. We compare the proposed model with the state-of-the-art models and various DL models, such as Ensemble2LSTM (Ens2-LSTM), bidirectional GRUs (Bi-GRU), and traditional machine learning (ML) models, such as support vector machine (SVM). Our experimental results based on the center for advanced studies in adaptive system and ARAS public data sets show that our model significantly outperforms the existing models for complex activity recognition of multiple residents.

Indoor localization is a crucial component of IoT applications in many areas, such as healthcare, energy management, and security control. Passive infrared (PIR) sensor has been employed for a location estimation due to its cost effectiveness, low power consumption, and low electromagnetic interference. Compared with its binary output, PIR analog output which is an output voltage generated by a PIR sensor when its sensing elements detect changes in temperature in an environment can provide more information regarding a person’s location. However, only a few works focus on using analog signals for location estimation. During the past several years, deep learning approaches have emerged and achieved outstanding results in many applications. In this article, we harness the power of deep learning and propose a deep CNN-LSTM architecture for PIR-based indoor location estimation. In our architecture, an upper CNN network can extract features from PIR analog output automatically while a lower LSTM network can learn temporal dependencies between the extracted features. To evaluate the feasibility and performance of our proposed method, we conduct four different sets of experiments. Our results show that the proposed method can efficiently handle complex cases and can achieve the mean distance error of 0.23 m, and 80% of distance errors are within 0.4 m.

Multiresident activity recognition (AR), which has become a popular research field in smart environments, aims to recognize the activities of multiple residents based on data collected from various types of sensors, and sensor events segmentation is an important technique for enhancing the performance of AR. While quite some segmentation methods have been proposed for the single person setting, few studies have been done for the multiresident setting. In this article, we first evaluate the baseline and the state-of-the-art segmentation methods using the popular multiresident data set CASAS, to confirm that the performance of multiresident AR can be improved by applying segmentation techniques; we then propose a novel Hybrid fuzzy c-means (FCM) change point detection (CPD)-based segmentation method that can further enhance the performance of multiresident AR. We combine a FCM method with a CPD-based method for sensor event segmentation. The FCM method is used to classify the sensor events in terms of sensor locations, and then the CPD technique is used to probe the transition actions to determine the segmentation sequence. Our experimental results show that the proposed method significantly improves the performance of multiresident AR in comparison with the baseline and state-of-the-art classification methods.

Summary Process view technologies allow organizations to create different granularity levels of abstraction of their business processes, therefore enabling a more effective business process management, analysis, interoperation, and privacy controls. Existing research proposed view construction and abstraction techniques for block‐based (ie, BPEL) and graph‐based (ie, BPMN) process models. However, the existing techniques treat each type of the two types of models separately. Especially, this brings in challenges for achieving a consistent process view for a BPEL model that derives from a BPMN model. In this paper, we propose a unified framework, namely UniFlexView, for supporting automatic and consistent process view construction. With our framework, process modelers can use our proposed View Definition Language to specify their view construction requirements disregarding the types of process models. Our UniFlexView's system prototype has been developed as a proof of concept and demonstration of the usability and feasibility of our framework.

In recent years, artifact-centric business process modeling is gaining momentum with its improved flexibility and extensibility. In order to support the rapid translation of the traditional activity-centric processes into this new type of processes, this paper proposes a novel approach to automatically transforming an activity-centric process model into a group of lifecycles of artifacts and their interactions, which represent the behavior of its corresponding artifact-centric process model. Algorithms on translating an activity-centric process model into a tree model, finding the dependencies between two object/artifact states based on the tree model, and synthesizing the lifecycles of artifacts have been proposed. Throughout the paper, we illustrate our approach with an order processing running scenario.