Global time aware conditioning improves pediatric sleep data analysis
Global Time-Aware Conditioning (GTC): A Mechanism for Time-Aware Pediatric Sleep Representation Learning
Machine Learning
Summary
Sleep changes throughout the night, but current computer methods usually look only at short parts of sleep without considering when those parts happen. The paper introduces a new way called Global Time-Aware Conditioning (GTC) that includes information about the exact time during sleep in the analysis. The authors added GTC to several existing methods and tested it on different tasks involving children's sleep data. They found that GTC helped the methods work better, especially when using data from different groups of children.
What this means in practice
- •For pediatric sleep clinicians: Assist in analyzing complex pediatric sleep recordings by incorporating timing context for better diagnosis and monitoring.
- •For medical device developers: Enhance automated pediatric sleep monitoring devices with time-aware models that improve detection accuracy across diverse patient groups.$Commercial implications: Enables more accurate sleep monitoring products for children by integrating global time context into existing algorithms.
- •For wearable device manufacturers: Develop more reliable sleep tracking wearables by using time-based conditioning of sleep data representations to improve sleep stage identification.$Commercial implications: Allows production of advanced consumer wearables delivering improved pediatric sleep insights via time-aware algorithms.
Authors
Saurav Raj Pandey, Harlin Lee
Abstract
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.