konstantinosbarmpas / motor-imagery-eeg-dynamic-convolutions

[TNSRE] IEEE Transactions on Neural Systems and Rehabilitation Engineering

Improving Generalization of CNN-based Motor-Imagery EEG Decoders via Dynamic Convolutions

IEEE Transactions on Neural Systems and Rehabilitation Engineering (TNSRE)

Published in IEEE Transactions on Neural Systems and Rehabilitation Engineering (TNSRE), 2023 - Paper

Authors: Konstantinos Barmpas, Yannis Panagakis, Stylianos Bakas, Dimitrios Adamos, Nikolaos Laskaris and Stefanos Zafeiriou

---

Deep Convolutional Neural Networks (CNNs) have recently demonstrated impressive results in electroencephalogram (EEG) decoding for several Brain-Computer Interface (BCI) paradigms, including Motor-Imagery (MI). However, neurophysiological processes underpinning EEG signals vary across subjects causing covariate shifts in data distributions and hence hindering the generalization of deep models across subjects. In this paper, we aim to address the challenge of inter-subject variability in MI. To this end, we employ causal reasoning to characterize all possible distribution shifts in the MI task and propose a dynamic convolution framework to account for shifts caused by the inter-subject variability. Using publicly available MI datasets, we demonstrate improved generalization performance (up to 5%) across subjects in various MI tasks for four well-established deep architectures.

---