Spatiotemporal Convolutions on EEG signal -- A Representation Learning Perspective on Efficient and Explainable EEG Classification with Convolutional Neural Nets

2026-05-05

AI Summary

This work addresses the limitations of conventional EEG classification models that rely on separate one-dimensional spatial and temporal convolutions, which struggle to efficiently capture joint spatiotemporal features and suffer from poor training efficiency in high-dimensional tasks. To overcome this, the authors propose a lightweight two-dimensional spatiotemporal convolutional architecture and systematically investigate the representational differences between 1D and 2D convolutions. Evaluated on a 22-channel motor imagery EEG task, the proposed method achieves comparable classification accuracy while significantly improving both training and inference efficiency. Through representational similarity analysis, the study further reveals—for the first time—that 1D and 2D models yield markedly distinct internal representational geometries, underscoring the critical role of architectural design in encoding multivariate neural signals.

Abstract

Classification of EEG signals using shallow Convolutional Neural Networks (CNNs) is a prevalent and successful approach across a variety of fields. Most of these models use independent one-dimensional (1D) convolutional layers along the spatial and temporal dimensions, which are concatenated without a non-linear activation layer between. In this paper, we investigate an alternative encoding that operates a bi-dimensional (2D) spatiotemporal convolution. While 2D convolutions are numerically identical to two concatenated 1D convolutions along the two dimensions, the impact on learning is still uncertain. We test 1D and 2D CNNs and a CNN+transformer hybrid model in a low-dimensional (3-channel) and a high-dimensional (22-channel) BCI motor imagery classification task. We observe that 2D convolutions significantly reduce training time in high-dimensional tasks while maintaining performance. We investigate the root of this improvement and find no difference in spectral feature importance. However, a clear pattern emerges in representational similarity across models: 1D and 2D models yield vastly different representational geometries. Overall, we suggest an improved model with a 2D convolutional layer for faster training and inference. We also highlight the importance of architecturally-driven encoding when processing complex multivariate signals, as reflected in internal representations rather than purely in performance metrics.

Problem

• EEG classification
• spatiotemporal convolution
• representation learning
• convolutional neural networks
• motor imagery

Innovation

• spatiotemporal convolution
• EEG classification
• representational geometry
• convolutional neural networks
• model interpretability