S4Sleep: Elucidating the design space of deep-learning-based sleep stage classification models

Tiezhi Wang, Nils Strodthoff

Abstract: Highlights•Developed systematic method for finding optimal time series annotation architectures.•Identified encoder–predictor models with S4 as best performing architecture.•Demonstrated strong performance using both raw time series and spectrogram inputs.•Validated S4’s effectiveness as signal encoder for both input representations.•Surpassed benchmarks on three major sleep datasets (Sleep EDF, MASS, SHHS).