Paper
Foundation model-based Bayesian imputation for wearables
ML4H 2026 poster (OpenReview 7SiaVXPHuQ, 2026-10-30)
TL;DR
Bayesian multiple imputation on top of the LSM-2 wearable FM: Monte Carlo dropout gives multiple imputations; Jensen-Shannon divergence between a day's missingness pattern and weekday-matched wear patterns rescales their spread. 73 participants; artificial MCAR, MAR, MNAR and combined missingness scored by CRPS. Heart-rate imputation error fell 33–48% under MAR, MCAR and combined, 12% under MNAR; no significant change for step counts.
- Form factor: consumer wearable
Results on ChatPPG leaderboards
All leaderboardsNo results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.
Tasks & signals
Signals
Keywords
How it works
Cite this paper
Hershkovich, L., Luan, H., Sim, I., & Dunn, J. (2026). Foundation model-based Bayesian imputation for wearables. ML4H 2026 poster. https://openreview.net/forum?id=7SiaVXPHuQ
Hershkovich, Leeor, Hongzhou Luan, Ida Sim, and Jessilyn Dunn. 2026. "Foundation model-based Bayesian imputation for wearables." ML4H 2026 poster. https://openreview.net/forum?id=7SiaVXPHuQ.
Hershkovich, L., Luan, H., Sim, I. and Dunn, J. (2026) 'Foundation model-based Bayesian imputation for wearables', ML4H 2026 poster. Available at: https://openreview.net/forum?id=7SiaVXPHuQ
Hershkovich, Leeor, et al. "Foundation model-based Bayesian imputation for wearables." ML4H 2026 poster, 2026. https://openreview.net/forum?id=7SiaVXPHuQ.
L. Hershkovich, H. Luan, I. Sim, and J. Dunn, "Foundation model-based Bayesian imputation for wearables," ML4H 2026 poster, 2026, [Online]. Available: https://openreview.net/forum?id=7SiaVXPHuQ.
@inproceedings{hershkovich2026foundation,
title = {{Foundation model-based Bayesian imputation for wearables}},
author = {Hershkovich, Leeor and Luan, Hongzhou and Sim, Ida and Dunn, Jessilyn},
year = {2026},
booktitle = {ML4H 2026 poster},
url = {https://openreview.net/forum?id=7SiaVXPHuQ}
}Added to the ChatPPG research hub