Uncertainty-Aware (Un)Supervised Few-Shot User Adaptation for On-Device Personalized Human Activity Recognition
For wearable HAR systems requiring on-device personalization, this work provides a lightweight and robust method that works with labeled, unlabeled, or no calibration data, addressing a practical deployment bottleneck.
This paper tackles the problem of personalizing HAR models to unseen users with limited calibration data. The proposed gradient-free framework, using closed-form Bayesian prototype estimation, achieves +2.76 to +33.44 pp macro-F1 improvement with supervised adaptation and +0.56 to +32.13 pp with unsupervised adaptation using only 3 seconds of calibration data per activity.
Sensor-based Human Activity Recognition (HAR) models often degrade on unseen users due to domain shifts caused by individual movement patterns and sensor placement. Practical wearable HAR systems therefore require personalization methods that are lightweight, applicable whether calibration data is labeled, unlabeled, or unavailable, and robust under limited calibration. We present a gradient-free framework that repurposes pretrained HAR classifiers as Prototypical Networks using using prior prototypes, which preserve zero-shot performance and regularize adaptation. For labeled calibration, we introduce closed-form Bayesian prototype estimation and extend the same principle to unlabeled calibration. With only 3 seconds of calibration data per activity (one shot), supervised adaptation improves macro-F1 by +2.76 to +33.44 percentage points across four datasets, while unsupervised adaptation improves by +0.56 to +32.13 points. Since adaptation requires only closed-form prototype updates, the framework enables efficient and robust on-device personalization of preexisting HAR classifiers.