Explainable machine learning for memory-related decoding via TabNet and non-linear features∗

Maxim Mametkulov, Abay Artykbayev, Darina Koishigarina, Amina Kenessova, Kamilla Razikhova, Taeho Kang, Christian Wallraven, Siamac Fazli

Abstract: In this study, we propose combining non-linear feature representations, namely Hurst Exponent, correlation dimension, and largest Lyapunov exponent, with TabNet, a novel attention-based neural network architecture, to perform EEG-based decoding of memory formation in single trials. Our results show that these combinations perform favourably when compared to current state-of-the-art approaches based on convolutional neural networks. Moreover, the interpretability of TabNet revealed that its feature selection for decision making is valid from a neurophysiological perspective, which is an advantage compared to other models.