A 3.4M-parameter foundation model pre-trained on step-count data alone achieves best AUROC on 20 of 21 health risk prediction tasks across multiple devices, regions, and diseases.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Physical activities enable scalable foundation modelling for broad-spectrum health prediction
A 3.4M-parameter foundation model pre-trained on step-count data alone achieves best AUROC on 20 of 21 health risk prediction tasks across multiple devices, regions, and diseases.