Uncertainty-Aware (Un)Supervised Few-Shot User Adaptation for On-Device Personalized Human Activity Recognition
arXiv:2606.04798v1 Announce Type: new 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 reg
The proliferation of wearable sensors and the increasing demand for personalized health and activity monitoring on edge devices necessitates robust and efficient on-device AI solutions for Human Activity Recognition (HAR).
This development addresses a key challenge in pervasive AI by enabling more accurate and adaptable on-device HAR, crucial for health, fitness, and human-computer interaction applications without relying on extensive labeled data or cloud processing.
The proposed gradient-free framework allows existing HAR classifiers to adapt quickly to new users with minimal data, improving personalization and reducing the computational and data demands for on-device AI deployments.
- · Wearable device manufacturers
- · Personalized health tech companies
- · Edge AI providers
- · Consumers of wearable technology
- · Cloud-dependent HAR solutions
- · Companies relying on extensive user calibration data
Improved accuracy and user satisfaction for personalized on-device human activity recognition applications becomes widely achievable.
The reduced need for labeled data and training on new users accelerates the deployment and scalability of intelligent wearable systems.
This efficiency could enable new forms of continuous, context-aware personalized AI services on small, low-power devices, extending into other domains beyond HAR.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG