Spiking Spatiotemporal Neural Architecture Search for EEG-Based Emotion Recognition

Spiking neural network (SNN) has the promising ability to take advantage of the spatiotemporal information from electroencephalogram (EEG) for emotion recognition. However, manually designing suitable SNN architectures needs considerable effort. In this article, we propose a novel and effective method, spiking spatiotemporal neural architecture search (SSTNAS), for EEG-based emotion recognition. SSTNAS exploits the discriminative spatial and temporal EEG features via spiking convolution neural network (SCNN) and spiking long short-term memory (SLSTM), respectively. Then, SSTNAS explores a proper SNN architecture for each task by investigating the spike activation patterns of pretrained networks based on genetic search, which is free of training. Experimental results on three public benchmark datasets, namely, FACED, DEAP, and DREAMER, demonstrate the superiority of the proposed method over the related state-of-the-art approaches.