SleepBand: Single-Source Domain Generalization for Sleep Staging via Physiologically Structured Spectral Modeling
2026-07-06 • Multimedia
MultimediaMachine Learning
AI summaryⓘ
The authors address the problem of training sleep stage classifiers on just one labeled dataset and making them work well on new, unseen datasets without extra labels. They created SleepBand, a method that uses brainwave patterns known to happen during sleep, captured by special filters, to focus on consistent features across datasets. Testing on five public datasets, their method performed very well and relied on meaningful brain signals instead of dataset-specific noise. This shows that including knowledge about sleep physiology helps models generalize better from single datasets.
Sleep stagingDomain generalizationSingle-source domain generalizationMorlet filter bankOscillatory priorsSlow wavesSleep spindlesNeurophysiologyLeave-one-domain-outPhysiology-aware inductive biases
Authors
Zhi Lu, Yang Hu, Yan Chen
Abstract
Generalizing sleep staging models to unseen datasets is challenging, and typical domain generalization (DG) methods often rely on multiple source domains or domain labels that are rarely available in practice. We tackle the stricter and more practical setting of single-source domain generalization: training on a single labeled source dataset, without domain labels or access to target data. We present SleepBand, a physiology-guided framework that embeds oscillatory priors via a learnable Morlet filter bank and a structured integration-and-recalibration pipeline. This anchors representations to domain-invariant sleep rhythms (e.g., slow waves, spindles), reducing reliance on dataset-specific artefacts. On five public datasets, SleepBand achieves state-of-the-art SDG performance and remains competitive under leave-one-domain-out (multi-source) DG. Analyses show that the learned filters align with canonical neurophysiology and that robustness stems from focusing on narrowband, physiologically meaningful cues. Our results suggest that principled, physiology-aware inductive biases are a promising path for robust single-domain sleep staging. Code is available at https://github.com/lzcn/sleep-band