EEG-ARNet: An Autoregressive Pre-training Model for Extracting EEG Features

EEG plays a crucial role in various domains due to its exceptional temporal resolution, yet its integration into deep learning models is impeded by challenges such as data scarcity and individual variability. This study presents EEG-ARNet, a novel autoregressive pre-training model designed to address these limitations by capturing the temporal dynamics of EEG signals. Through extensive experiments on the EEG-based visual classification dataset, EEG-ARNet demonstrates superior performance in terms of accuracy and Cohen's Kappa when compared to traditional methods, as well as autoencoder-based pre-trained models. The model's ability to extract rich EEG features through autoregressive learning is validated across multiple classification tasks, highlighting its potential in advancing EEG signal processing. Future directions will focus on incorporating spatial EEG information and expanding the model's application to brain-computer interfaces.