Track genomics, neuroscience and molecular biology.
Search papers, labs, and topics across Lattice.
This paper introduces Global Time-Aware Conditioning (GTC), a novel mechanism that augments local positional encodings in self-supervised learning (SSL) for sleep by conditioning latent representations on the global time-of-sleep for each input window. By explicitly modeling the temporal progression of sleep physiology, GTC addresses the limitations of fixed-duration window processing in existing SSL methods. Evaluation across three SSL frameworks and five clinically relevant tasks on pediatric sleep data demonstrates that GTC consistently improves F1 and AUPRC scores, even under cross-cohort transfer, underscoring the importance of global temporal context.
Explicitly conditioning sleep representations on global time-of-sleep dramatically boosts performance on challenging pediatric sleep tasks, even across different datasets and models.
Sleep physiology evolves throughout the night, with changes in sleep cycles, respiratory stability, and cardiovascular dynamics. However, self-supervised learning (SSL) methods for sleep typically process short, fixed-duration windows without explicitly modeling where each window occurs within the full sleep session. We introduce Global Time-Aware Conditioning (GTC), a mechanism that complements local positional encodings by conditioning latent representations on the global time-of-sleep of each input window. We integrate GTC into three representative SSL frameworks and evaluate them on five clinically relevant tasks with extreme class imbalance across two independent pediatric cohorts. GTC generally improves F1 and AUPRC across architectures and tasks, with gains persisting under cross-cohort transfer, highlighting the value of global temporal context for sleep representation learning.