Spatiotemporal Convolutions on EEG signal - A Representation Learning Perspective on Efficient and Explainable EEG Classification with Convolutional Neural Nets
1 st Laurits Dixen, 2 nd Stefan Heinrich, 3 rd Paolo Burelli · 2026
Read Spatiotemporal Convolutions on EEG signal - A Representation Learning Perspective on Efficient and Explainable EEG Classification with Convolutional Neural Nets by 1 st Laurits Dixen, 2 nd Stefan Heinrich, 3 rd Paolo Burelli (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-03874.txt.
Description
Laurits Dixen, Stefan Heinrich, and Paolo Burelli compare ways of convolving over EEG signals for classification with shallow convolutional networks. Most models convolve separately along the spatial and temporal dimensions. They then join the results, with no nonlinearity in between. The authors test an alternative: a single two-dimensional spatiotemporal convolution. On paper it is numerically identical to the two one-dimensional ones. Its effect on learning was still uncertain.
They compare one-dimensional and two-dimensional models, plus a convolutional-transformer hybrid, on brain-computer motor imagery tasks with three and twenty-two channels. Two-dimensional convolutions cut training time a lot on the high-dimensional task, while keeping performance. Spectral feature weight did not differ. But a clear pattern emerged in representational similarity: the two encodings learn vastly different geometries. The authors suggest the two-dimensional design. It trains and runs faster.
Catalog record
| CREATOR | 1 st Laurits Dixen, 2 nd Stefan Heinrich, 3 rd Paolo Burelli |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.03874