Reproduce “Learning transferable human physiology from two million hours of sleep with SleepFM-2” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rahul Thapa, Christopher Sun, William Theodor Lehn-Schioler, Sophia Claire Kivelson, Umaer Hanif, Hyatt Moore IV, Harrison G. Zhang, Hafsa Ahmed and 9 more
SleepFM-2 uses a frozen model trained on overnight polysomnography to combine brain, eye, heart, muscle, breathing and oxygen signals for clinical prediction. It represents one-second channel patches jointly, then task-specific heads reuse that representation. It identified 188 incident diagnoses meeting the authors’ criteria in both cohorts, including an unseen health system; these retrospective risk rankings do not establish disease onset.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.