Machine Learning for Health

Learning from wearable, physiological, and omics data to model human health.

Medicine observes health through incomplete, noisy measurements: wearables, laboratory tests, imaging, and clinical records. I build models that connect these signals to underlying physiology and stay reliable outside controlled studies, where compliance, device effects, and domain shift dominate. Examples include modeling heart rate and oxygen consumption from consumer wearables, sleep phenotyping in autism, and single-cell perturbation response. The longer-term goal is to determine what evidence about a person is enough, and which additional measurement is worth acquiring.

Selected work

* Equally-contributing first author, † Equally-contributing senior author

2026

  1. vedula2026sleep.png
    MLHC2026
    Real-world sleep phenotyping in autism: compliance, domain shift, and scalable wearable biomarkers
    S. Vedula, A. Kim, M. Kumar, O. Troyanskaya, G. Sapiro
    Proc. Machine Learning for Healthcare Conference (MLHC)
  2. ICLR Workshop
    ICLR Workshop2026
    On the role of drug representations in single-cell perturbation modeling
    S. Vedula*, M. Pegoraro*, L. Halika, M. Danziger, A. Bronstein, M. Rosen-Zvi
    ICLR 2026 Learning Meaningful Representations of Life (LMRL) Workshop

2025

  1. bioRxiv
    bioRxiv2025
    An Evidence-Grounded Research Assistant for Functional Genomics and Drug Target Assessment
    K. Sokolova, D. Kosenkov, K. Nallamotu, S. Vedula, D. Sokolov, G. Sapiro, O. Troyanskaya
    bioRxiv
    HTML
  2. gahtan2025lab.png
    ICMI2025
    From Lab to Wrist: Bridging Metabolic Monitoring and Consumer Wearables for Heart Rate and Oxygen Consumption Modeling
    B. Gahtan, S. Vedula, G. Samuelly Leichtag, E. Kodesh, A. Bronstein
    Proceedings of the 27th International Conference on Multimodal Interaction

← All research areas