Human Activity Recognition (HAR) is a cornerstone of precision health, enabling personalized health management through the continuous monitoring of daily activities and behavioral patterns. Many existing HAR methods prioritize advancing feature extraction architectures while often overlooking the detrimental impact of inherent data noise. This oversight can compromise feature integrity and, consequently, degrade recognition accuracy. To address this limitation, we introduce the Multimodal Uncertainty Fusion Network (MUFNet), a novel framework that integrates skeleton and RGB data streams. A key innovation of MUFNet is the Feature Uncertainty Enhancement Module (FUEM), which models feature reliability by learning a class-aligned Gaussian distribution. This allows the model to stochastically identify and amplify robust feature channels, thereby enhancing feature resilience to noise. Concurrently, a Dynamic Temporal Convolution (DTC) module employs learnable continuous offsets to adeptly capture actions of varying temporal dynamics, including those that are rapid or irregular. Our approach fuses features from multiple skeleton modalities (3D joint, 3D bone, and 2D joint) with spatiotemporal region-of-interest (ST-ROI) features extracted from RGB frames, where the latter is guided by the 2D skeleton structure. Comprehensive evaluations on three public multimodal HAR benchmarks— NTU RGB+D 60, NTU RGB+D 120, and Northwestern-UCLA—validate the superiority of MUFNet. Our model consistently outperforms state-of-the-art methods, achieving accuracies of 95.6% (X-Sub) / 98.9% (X-View) on NTU RGB+D 60, 91.3% (X-Sub) / 92.9% (X-Set) on NTU RGB+D 120, and 95.3% on Northwestern-UCLA. The source code is available at: https://github.com/ShiZhiLi7/MUFNet.

Quantum Dempster-Shafer theory (QDST) derives a quantum mass function (QMF), a fuzzy metric obtained from multiple information sources based on quantum interference. In general, QMF effectively represents and processes uncertain information, but managing conflicts among multiple QMFs remains challenging. To address this issue, we propose a novel quantum conflict indicator (QCI) within the QDST framework. It is the first metric satisfying ideal conflict measurement properties, including non-negativity, symmetry, boundedness, extreme consistency, and insensitivity to refinement. Based on QCI, a novel quantum conflict fusion method (QCI-Fusion) is introduced to fuse highly conflicting QMFs. Moreover, traditional methods, including QCI-Fusion, typically constructs the Quantum Frame of Discernment (QFoD) based on predicted labels, which makes it difficult to cover unseen classes. Therefore, a new decision architecture, QCI-Decision, is proposed for unsupervised detection that rejects out-of-distribution (OOD) samples while maintaining in-distribution (ID) classification. Experimental results show that the classification accuracy of QCI-Decision deviates from the original model predictions by at most 1.49%. Meanwhile, compared with the latest OOD detection methods, QCI-Decision improves the Area Under the Receiver Operating Characteristic Curve (AUC) by up to 0.6% and reduces the False Positive Rate at 95% True Negative Rate (FPR) by up to 1.63%. Moreover, compared with QCI-Fusion, QCI-Decision achieves approximately threefold faster fusion speed with negligible performance degradation, offering a promising solution for open-world quantum information decision.

The brain-computer interface (BCI) system facilitates efficient communication and control, with Electroencephalography (EEG) signals as a vital component. Traditional EEG signal classification, based on static deep-learning models, presents a challenge when new classes of the subject's brain activity emerge. The goal is to develop a model that can recognize new few-shot classes while preserving its ability to discriminate between existing ones. This scenario is referred to as Few-Shot Class-Incremental Learning (FSCIL). This work introduces IncrementEEG, a novel framework meticulously designed to tackle the distinct challenges of FSCIL in EEG-based brain activity classification, focusing specifically on emotion recognition and steady-state visual evoked potential (SSVEP). Our work analyzes the role of additive angular margin loss in improving the model's discrimination capabilities. The proposed method is designed to demonstrate robustness in open-world conditions and adaptability to new tasks. Furthermore, we introduce a prototype refinement module comprising a prototype augmentation block and an update block. The prototype augmentation block in the deep feature space preserves the decision boundary for prior tasks, and the prototype update block utilizes a shared embedding space to compute the relation matrix for bootstrapping prototype updates. Extensive experiments conducted across multiple datasets show the superior performance of the IncrementEEG framework compared to state-of-the-art methods. The proposed method advances FSCIL brain activity classification, offering promising potential for applications in Brain-Computer Interface systems.

Combining brain-computer interface (BCI) technology with the Internet of Things (IoT) for practical motor rehabilitation applications is a critical research direction aimed at enhancing the decoding performance of BCIs and their clinical application value. Currently, BCIs for single rehabilitation movements have started to be applied in clinical settings. However, in the field of electroencephalography (EEG) signal decoding, we note that research specifically focused on motor-rest classification within single motor imagery (MI) tasks remains relatively limited. To address this gap, we propose the Bidirectional Channel Attention and Mutual-Cross-Attention Temporal Feature Extraction Network (Bi-ACTCNet), adapted for operation within IoT environments. This network excels in performing Motor-Rest classification across three datasets, yielding better outcomes compared to existing models. A key innovation of Bi-ACTCNet lies in its bidirectional temporal feature extraction method, which reveals that reverse temporal signals in EEG data contain additional features that are advantageous for improving model accuracy. Additionally, the model improves the Efficient Channel Attention (ECA) block by introducing the Efficient Temporal Channel Attention (ETCA) block, and for the first time, it utilizes Mutual-Cross-Attention (MCA) to fuse features from the bidirectional branches. Ablation experiments have verified the effectiveness of each block. To further substantiate the model’s effectiveness and generalizability, we executed 2-class classification experiments using the BCI Competition IV-2a dataset, achieved excellent performance.

In recent years, the utilization of wearable sensors for Human Activity Recognition (HAR) has garnered significant interest in the fields of medical health monitoring and sports management. However, HAR often suffer the poor generalization from the insufficient labeled data for complex activities. To address this issue, the novel Transfer Component Analysis-Bidirectional Long Short-Term Memory network (TCA-BiLSTM) with the fused attention mechanism is presented in this paper. Specifically, TCA-BiLSTM first leverages the Maximum Mean Difference (MMD) within the Reproducing Kernel Hilbert Space (RKHS) to learn transfer components for sensor-based HAR. These derived transfer components align the data collected from sensors deployed on different body parts, facilitating the mapping of cross-domain HAR data. Then, the two-layer BiLSTM with the novel fused attention mechanism is given to classify the unseen activities, which aims to capture the multi-granularity activity information after the TCA-based domain adaptation. To evaluate the effectiveness of TCA-BiLSTM, a series of experiments were conducted using the DSADS and PAMAP2 datasets. The results demonstrate that TCA-BiLSTM outperforms the state-of-art methods such as DSAN and FNet, achieving performance improvements of 6.1% and 2.5%, respectively.

Decoding human visual representations from brain activity data is a challenging but arguably essential task with an understanding of the real world and the human visual system. However, decoding semantically similar visual representations from brain recordings is difficult, especially for electroencephalography (EEG), which has excellent temporal resolution but suffers from spatial precision. Prevailing methods mainly focus on matching brain activity data with corresponding stimuli-responses using contrastive learning. They rely on massive and high-quality paired data and omit semantically aligned modalities distributed in distinct regions of the latent space. This paper proposes a novel Multimodal Bidirectional Cycle Consistency (MB2C) framework for learning robust visual neural representations. Specifically, we utilize dual-GAN to generate modality-related features and inversely translate back to the corresponding semantic latent space to close the modality gap and guarantee that embeddings from different modalities with similar semantics are in the same region of representation space. We perform zero-shot tasks on the ThingsEEG dataset. Additionally, we conduct EEG classification and image reconstruction on both the ThingsEEG and EEGCVPR40 datasets, achieving state-of-the-art performance compared to other baselines.

Credit card fraud detection (CCFD) is an important issue concerned by financial institutions. Existing methods generally employ aggregated or raw features as their representations to train their detection models. Yet such features tend to fall short of effectively exposing the characteristics of various frauds. In this work, we propose a spatial-temporal gated network (STGN) to automatically learn new informative transactional representations containing users’ transactional behavioral information for CCFD. A gated recurrent neural net unit is specifically constructed with a time-aware gate and location-aware gate to extract users’ spatial and temporal transactional behaviors. A spatial-temporal attention module is designed to expose the transaction motive of users in their historical transactional behaviors, which allows the proposed model to better extract the fraudulent characteristics from successive transactions with time and location information. A representation interaction module is offered to make rational decisions and learn compositive transactional representations. A real-world transaction dataset is used in experiments to verify the efficacy of the learned new representations. The results demonstrate that our proposed model outperforms the state-of-the-art ones, thus greatly advancing the field of CCFD.Note to Practitioners—The features of transaction records reflect the characteristics of users’ transactional behaviors. Therefore, effective features are critical for accurate CCFD. However, fraudsters often pretend to be legitimate users during transactions to deceive the CCFD system. As a result, fraudulent behaviors become concealed within legitimate ones, signifying that original features are inadequate for accurate CCFD. Thus, it is imperative for researchers and practitioners to extract new features that can well expose fraud characteristics. While existing methods employing some transaction aggregation strategies can spot certain fraudulent behaviors, they fail to clearly cluster all the anomalous behaviors and distinguish them from legitimate behaviors. Therefore, this work is driven by the urgent demand to extract new informative features for CCFD. Its primary focus is to unveil the aggregation of fraudulent transactional behaviors from both temporal and spatial perspectives, enabling more accurate CCFD. Specifically, this work introduces a new STGN model that automatically learns new transactional representations incorporating users’ transactional behavioral information for CCFD. By comprehensively considering the time interval and location interval of consecutive user transactions, we thoroughly reveal the temporal and spatial aggregation of fraudulent behavior, which provides valuable insights for CCFD practitioners: 1) employing features that integrate the behavioral characteristics of fraudsters instead of the original features can enhance the model’s capability to identify frauds, and 2) taking into account the time and location intervals of users’ consecutive historical transactions can better uncover the behavioral characteristics of fraudsters.

In recent years, wearable sensor-based human activity recognition (HAR) is becoming more and more attractive, especially in health monitoring and sports management. However, in order to obtain high-quality HAR, it is often necessary to get sufficient labeled activity data, which is very difficult, time-consuming, and costly in a natural environment. To tackle this problem, multisource domain adaptation (DA) is a promising method that aims to learn enough multisource prior knowledge from labeled activity data, and then transfer this learned knowledge to the target unlabeled dataset. Thus, this article presents a novel multisource weighted DA with evidential reasoning (w-MSDAER) for HAR, which can effectively utilize complementary knowledge between multiple sources. Specifically, we first use the strategy of distribution alignment to learn local domain-invariant classifiers based on multisource domains. And then the reliabilities of these derived classifiers are comprehensively evaluated according to the belief function based technique for order preference by similarity to ideal solution (BF-TOPSIS). Finally, the discounting fusion method is used to fuse the local classification results. Comprehensive experiments are conducted on two open-source datasets, and the results show that the proposed w-MSDAER significantly outperforms other state-of-art methods.

In recent years, spiking neural networks (SNNs), which originated from the theoretical basis of neuroscience, have attracted neuromorphic computing and brain-like computing due to their advantages, such as neural dynamics and coding mechanism, which are similar to biological neurons. SNNs have become one of the mainstream frameworks in the field of brain-like computing. However, most of the Leaky Integrate-and-Fire (LIF) neuron models currently used by SNNs based on direct training of backpropagation (BP) do not consider the changes in the recurrent connections and the dynamic strength of neuron connections over time. This study presented the LIF neuron model with recurrent connections and a method for dynamically changing the presynaptic currents. Recurrent LIF neurons have an additional cyclic connection compared with classic LIF neurons. Their postsynaptic current stimulates a change in membrane potential at the next time point. Their dynamics were more similar to the activities of biological neurons. We also proposed an efficient and flexible BP training method for recurrent LIF neurons. On the basis of the above methods, we proposed the recurrent SNN with dynamic presynaptic currents based on backpropagation (RDS-BP). We test the proposed RDS-BP on three image data sets (MNIST, Fashion-MNIST and CIFAR-10) and two text data sets (IMDB and TREC). The results showed that the performance of RDS-BP not only exceeded the naive SNN models based on BP but also exceeded the SNN methods proposed in previous studies in recent years, which had excellent performance in previous experiments. Our work provides a new LIF neuron model with a recurrent connection and dynamic presynaptic current and a BP training arrangement for the proposed neuron, which could merit developments with neuromorphic and brain-like computing.

Summary Motor imagery (MI) is an important control paradigm in the field of brain‐computer interface (BCI), which enables the recognition of personal intention. So far, numerous methods have been designed to classify EEG signal features for MI task. However, deep neural networks have been seldom applied to analyze EEG signals. In this study, two novel kinds of deep learning schemes based on convolutional neural networks (CNN) and Long Short‐Term Memory (LSTM) were proposed for MI‐classification. The frequency domain representations of EEG signals were obtained using short time Fourier transform (STFT) to train models. Classification results were compared between conventional algorithm, CNN, and LSTM models. Compared with two other methods, CNN algorithms had shown better performance. These conclusions verified that CNN method was promising for MI‐based BCIs.

There have been many attempts to design brain-computer interfaces (BCIs) for wheelchair control based on steady state visual evoked potential (SSVEP), event-related desynchronization/synchronization (ERD/ERS) during motor imagery (MI) tasks, P300 evoked potential, and some hybrid signals. However, those BCI systems cannot implement the wheelchair navigation flexibly and effectively. In this paper, we propose a hybrid BCI scheme based on two-class MI and four-class SSVEP tasks. It cannot only provide multi-degree control for its user, but also allow the user implement the different types of commands in parallel. In order for the subject to learn the hybrid mental strategies effectively, we design a visual and auditory cues and feedback-based training paradigm. Furthermore, an algorithm based on entropy of classification probabilities is proposed to detect intentional control (IC) state for hybrid tasks, and ensure that multi-degree control commands are accurately and quickly generated. The experiment results attest to the efficiency and flexibility of the hybrid BCI for wheelchair control in the real-world.