Uncertainty-Aware (Un)Supervised Few-Shot User Adaptation for On-Device Personalized Human Activity Recognition
2026-06-03 • Machine Learning
Machine Learning
AI summaryⓘ
The authors address a problem where activity recognition models struggle when used by new people because everyone moves differently and wears sensors slightly differently. They propose a simple method to personalize these models quickly using very little new data, even if the new data isn’t labeled or is unlabeled. Their approach, based on a type of model called Prototypical Networks, updates the system efficiently on a device without complicated calculations. Experiments show their method improves recognition accuracy significantly across several datasets with just a few seconds of new data.
Human Activity Recognitiondomain shiftwearable sensorspersonalizationPrototypical Networkszero-shot learningBayesian estimationunsupervised adaptationmacro-F1 score
Authors
Maximilian Burzer, Till Riedel, Michael Beigl, Tobias Röddiger
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.