Reaching the ceiling? Empirical scaling behaviour for deep EEG pathology classification

Ann-Kathrin Kiessner, Robin Tibor Schirrmeister, Joschka Boedecker, Tonio Ball

Abstract: Highlights•We investigated the impact of model and dataset size on EEG pathology classification.•Test performance improves as a saturating power-law with model and dataset size.•Empirically observed accuracies of current EEG pathology ConvNets saturate near 87%.•Classification accuracies may limited by clinical label noise.