EdgeSense: Federated Open-Set Activity Recognition with Privacy-Aware Novelty Detection

Rong Fu, Anqi Zhao, weizhi Tang, Simon James Fong

Keywords: Federated Learning, Open-Set Recognition, Human Activity Recognition, Wi-Fi CSI, Edge Computing, Novelty Detection, Privacy-Preserving AI, Streaming Data

TL;DR: We propose a federated edge-based framework for open-set human activity recognition with privacy-aware novelty detection using Wi-Fi CSI.

Abstract: This paper introduces a federated framework for open-set human activity recognition in medical Internet of Things environments using Wi-Fi channel state information. The proposed system integrates lightweight novelty detection with edge-based incremental learning, enabling real-time identification of previously unseen activity patterns while preserving privacy. By decoupling recognition and novelty detection, the architecture supports dynamic label space expansion and distributed consensus validation across resource-constrained devices. The framework operates without transmitting raw data, ensuring compliance with privacy standards. Empirical evaluations in simulated clinical settings demonstrate robust performance under evolving behavioral conditions, establishing a scalable and privacy-conscious solution for adaptive activity monitoring.

Primary Area: learning on time series and dynamical systems

Code Of Ethics: true

Submission Guidelines: true

Anonymous Url: true

No Acknowledgement Section: true

Submission Number: 7326