Sleep stage classification is essential for sleep assessment and disease diagnosis. Although previous attempts to classify sleep stages have achieved high classification performance, several challenges remain open: 1) How to effectively utilize time-varying spatial and temporal features from multi-channel brain signals remains challenging. Prior works have not been able to fully utilize the spatial topological information among brain regions. 2) Due to the many differences found in individual biological signals, how to overcome the differences of subjects and improve the generalization of deep neural networks is important. 3) Most deep learning methods ignore the interpretability of the model to the brain. To address the above challenges, we propose a multi-view spatial-temporal graph convolutional networks (MSTGCN) with domain generalization for sleep stage classification. Specifically, we construct two brain view graphs for MSTGCN based on the functional connectivity and physical distance proximity of the brain regions. The MSTGCN consists of graph convolutions for extracting spatial features and temporal convolutions for capturing the transition rules among sleep stages. In addition, attention mechanism is employed for capturing the most relevant spatial-temporal information for sleep stage classification. Finally, domain generalization and MSTGCN are integrated into a unified framework to extract subject-invariant sleep features. Experiments on two public datasets demonstrate that the proposed model outperforms the state-of-the-art baselines.

Sleep stage classification is critical for diagnosing sleep quality. While deep neural networks are becoming popular for automatic sleep stage classification with supervised learning, large-scale labeled datasets are still hard to acquire. Recently, self-supervised learning (SSL) has become one of the most prominent approaches to alleviate the burden of labeling works. However, existing SSL methods are mainly designed for non-temporally correlated data and only learning representations from an instance level. Hence, the objective of this paper is to learn robust and generalizable representations for physiological signals with self-supervised learning. Specifically, we make the following contributions: (1) we propose a novel co-training scheme by exploiting complementary information from multiple views (time view and frequency view) of physiological signals to mine more positive samples, which overcomes the drawback of popular InfoNCE and achieves semantic-level representation learning; (2) we extend our framework with a memory module, implemented by a queue and a moving-averaged encoder, to enlarge the pool of negative candidates and keep the up-to-date representation; (3) extensive experiments conducted on sleep stage classification demonstrate state-of-the-art performance compared with SSL baselines, achieving 71.6% and 57.9% accuracies on two sleep datasets, SleepEDF and ISRUC respectively. The code is publicly available at https://github.com/larryshaw0079/CoSleep .

In industrial facilities or IT systems, there are lots of multivariate time series generated from various metrics. Anomaly detection in multivariate time series is of great importance in applications such as fault diagnosis and root cause discovery. Recently, some unsupervised methods have made great progress in this task, especially reconstruction architecture of AutoEncoders (AEs), learning normal distribution and producing significant error for anomalies. Although AEs can reconstruct subtle abnormal patterns well with the powerful generalization ability, it also leads to a high false negative. Moreover, these AE-based models ignore the dependence among variables at different time scales. In this paper, we propose a novel anomaly detection framework named Multi-scale wavElet Graph Autoencoder (MEGA). The main idea is to integrate discrete wavelet transform into AE to decompose multivariate time series into multi-frequency components and then reconstruct them, thereby highlighting various anomalies in specific frequency bands. Meanwhile, for anomalies caused by the changes of inter-variable dependence, we introduce a dynamic graph module to capture such dependence on the decomposed multi-scale frequency components. Experiments have been carried on three public multivariate time series anomaly detection datasets. The results demonstrate that the MEGA outperforms the state-of-the-art baselines.

Emotion recognition has become an important area in affective computing. Emotion recognition based on multichannel electroencephalogram (EEG) signals has gradually become popular in recent years. However, on one hand, how to make full use of different EEG features and the discriminative local patterns among the features for various emotions is challenging. Existing methods ignore the complementarity among the spatial–spectral–temporal features and discriminative local patterns in all features, which limits the classification performance. On the other hand, when dealing with cross-subject emotion recognition, existing transfer learning (TL) methods need a lot of training data. At the same time, it is extremely expensive and time-consuming to collect the labeled EEG data, which is not conducive to the wide application of emotion recognition models for new subjects. To solve the above challenges, we propose a novel spatial–spectral–temporal-based attention 3-D dense network (SST-Net) with meta-learning, named MetaEmotionNet, for emotion recognition. Specifically, MetaEmotionNet integrates the spatial–spectral–temporal features simultaneously in a unified network framework through two-stream fusion. At the same time, the 3-D attention mechanism can adaptively explore discriminative local patterns. In addition, a meta-learning algorithm is applied to reduce dependence on training data. Experiments demonstrate that the MetaEmotionNet is superior to the baseline models on two benchmark datasets.