EEG Conformer: Convolutional Transformer for EEG Decoding and Visualization

Abstract

Due to the limited perceptual field, convolutional neural networks (CNN) only extract local temporal features and may fail to capture long-term dependencies for EEG decoding. In this paper, we propose a compact Convolutional Transformer, named EEG Conformer, to encapsulate local and global features in a unified EEG classification framework. Specifically, the convolution module learns the low-level local features throughout the one-dimensional temporal and spatial convolution layers. The self-attention module is straightforwardly connected to extract the global correlation within the local temporal features. Subsequently, the simple classifier module based on fullyconnected layers is followed to predict the categories for EEG signals. To enhance interpretability, we also devise a visualization strategy to project the class activation mapping onto the brain topography. Finally, we have conducted extensive experiments to evaluate our method on three public datasets in EEGbased motor imagery and emotion recognition paradigms. The experimental results show that our method achieves state-of-theart performance and has great potential to be a new baseline for general EEG decoding. The code has been released in https://github.com/eeyhsong/EEG-Conformer.

Adaptive transfer learning for EEG motor imagery classification with deep Convolutional Neural Network

“…Recently, deep learning (DL) methods such as Convolutional Neural Networks (CNNs) [5], Recurrent Neural Networks (RNNs) [6], and Transformers [7] have received increasing interest as a mean for classifying EEG signals [8]- [10]. At present, in addition to conventional CNNs or RNNs, DL of EEG incorporates many other technologies [11]- [15].…”

Representative-Based Cold Start for Adaptive SSVEP-BCI

“…Finally, when the pre-trained model is fitted on UI data with stronger transferability, fewer adjustments are needed to achieve rapid convergence to the ideal UD model. It is still possible that a large-scale unified DNN algorithm can be trained to render a more effective pre-trained model [28], or the linear combination of the representatives can be used to further boost performance [7]. However, the authors still want to underline that it is feasible build an eligible pre-trained model with state-of-the-art algorithms, if inter-data transferability is respected.…”

Algorithm contest of motor imagery BCI in the World Robot Contest 2022: A survey

“…Additionally, some applications of attention-based DL networks in MI have been developed in recent years. Song et al [62] proposed EEG Conformer, a convolutional transformer for EEG decoding and visualization. Liu et al [63] proposed TCACNet, a temporal and channel attention convolutional network for MI-EEG classification.…”