Foundation models for movement data: Are they ready for prime-time?

Summary

This paper presents what the authors describe as the first comprehensive evaluation of open-source accelerometer foundation models against supervised baselines. Foundation models trained on large-scale accelerometer corpora have been proposed as general-purpose feature extractors for health monitoring, but the authors note that systematic evidence for their advantages has been lacking. They evaluate four open accelerometer foundation models across 19 tasks spanning three domains: activity recognition including activities of daily living, clinical monitoring, and physiological inference.