Towards a General Intelligence and Interface for Wearable Health Data

2026-05-21

AI Summary

This work addresses the challenge of effectively characterizing personalized health states from low-level wearable sensor signals, which is hindered by substantial inter-individual variability and scarce labeled data. To this end, the authors develop a foundation model for wearable health data, pretrained on over 5 million participants and more than one trillion minutes of unlabeled time-series data. By synergistically scaling model and data size, leveraging few-shot transfer learning, and employing an embedding-driven predictive architecture—augmented for the first time with a large language model (LLM) agent to automatically optimize prediction heads—the approach enables context-aware, generative estimation of health metrics. Evaluated across 35 tasks spanning cardiovascular, metabolic, sleep, and mental health domains, the method demonstrates consistent performance gains. In assessments by 1,860 clinicians, the resulting personal health agent exhibited responses that were more relevant, safer, and better grounded in contextual understanding.

Abstract

While ubiquitous wearable sensors capture a wealth of behavioral and physiological information, effectively transforming these signals into personalized health insights is challenging. Specifically, converting low-level sensor data into representations capable of characterizing higher-level states is difficult due to high phenotypic diversity and variation in individual baseline health, physiology, and lifestyle factors. Moreover, collecting wearable data paired with health outcome annotations is laborious and expensive, and retrospective annotation remains practically unfeasible, contributing to a scarcity of data with high-quality labels. To overcome these limitations, we propose a foundation model for wearable health that is pretrained on more than one trillion minutes of unlabeled sensor signals drawn from a large cohort of five million participants. We demonstrate that the joint scaling of model capacity and pretraining data volume leads to systematic improvements in performance, as evaluated on a diverse set of 35 health prediction tasks, spanning cardiovascular, metabolic, sleep, and mental health, as well as lifestyle choices and demographic factors. We find that this population scale representation unlocks label-efficient few-shot learning and generative capabilities for robust daily metric estimation. To further leverage this learned representation, we deploy a classroom of LLM agents to autonomously search the space of downstream predictive heads built on the model embeddings, showing broad performance improvements that increase with LLM model capacity. Finally, we show how integrating these downstream predictors into a Personal Health Agent can support model responses that are more relevant, contextually aware, and safe, and we validate this via 1,860 ratings from a cohort of clinicians.

Problem

• wearable health data
• personalized health insights
• phenotypic diversity
• label scarcity
• health state representation

Innovation

• foundation model
• wearable health data
• few-shot learning
• LLM agents
• Personal Health Agent