Uncertainty-Aware (Un)Supervised Few-Shot User Adaptation for On-Device Personalized Human Activity Recognition

2026-06-03

AI Summary

This work addresses the performance degradation of sensor-based human activity recognition (HAR) models on new users due to inter-individual variability and differences in sensor placement. To tackle this challenge, the authors propose a lightweight, gradient-free personalization framework that transforms a pre-trained HAR classifier into a prototypical network augmented with closed-form Bayesian prototype estimation. This approach enables supervised, unsupervised, and zero-shot adaptation using only three seconds of calibration data per activity class. Notably, it represents the first integration of prototypical networks with Bayesian prototype estimation for HAR user adaptation and incorporates an uncertainty-aware mechanism. Evaluated on four benchmark datasets, the method achieves macro F1-score improvements of 2.76–33.44 percentage points under supervised adaptation and 0.56–32.13 points under unsupervised adaptation, substantially outperforming existing approaches.

Abstract

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.

Problem

• Human Activity Recognition
• Domain Shift
• Few-Shot Adaptation
• On-Device Personalization
• Uncertainty-Aware

Innovation

• Uncertainty-Aware Adaptation
• Prototypical Networks
• Few-Shot Personalization
• On-Device Learning
• Bayesian Prototype Estimation