Scene-Adaptive Continual Learning for CSI-based Human Activity Recognition with Mixture of Experts

2026-05-07

AI Summary

This work addresses the performance degradation in cross-domain CSI-based human activity recognition (CSI-HAR) caused by domain shift, a challenge exacerbated by existing continual learning approaches that often incur high inference overhead, rely on large replay buffers, or suffer from poor scalability. To overcome these limitations, the authors propose SAMoE-C, a framework that formulates cross-domain CSI-HAR as a mixture-of-experts system. It employs an attention-driven semantic routing mechanism to adaptively activate only relevant experts for each input, enabling scene-aware inference. Coupled with clustering-guided expert design and a lightweight continual learning protocol leveraging a minimal replay buffer, SAMoE-C achieves near state-of-the-art accuracy across four real-world CSI datasets while substantially reducing both inference cost and training burden, thereby facilitating efficient and scalable deployment in practical scenarios.

Abstract

Channel state information (CSI)-based human activity recognition (HAR) is vulnerable to performance degradation under domain shifts across varying physical environments. Continual learning (CL) offers a principled way to learn new domains sequentially while preserving past knowledge, but existing CL solutions for CSI-based HAR scale poorly with accumulating domains, rely on a large replay buffer, or incur linearly growing inference cost. In this letter, we propose Scene-Adaptive Mixture of Experts with Clustered Specialists (SAMoE-C), which formulates cross-domain CSI-based HAR as a mixture-of-experts system that enables scene-specific adaptation, via an attention-based semantic router that activates only selected experts for each input. Moreover, we develop a novel training protocol, which requires only a tiny replay buffer for stabilizing domain discrimination of the router. Experimental results on a four-scene CSI dataset demonstrate that SAMoE-C approaches the state-of-the-art accuracy, while maintaining a significantly lower inference cost. By jointly combining modular experts, selective activation with router and a lightweight training protocol, SAMoE-C enables scalable cross-domain CSI-based HAR deployment with low training overhead and high computational efficiency in real-world settings.

Problem

• CSI-based HAR
• domain shift
• continual learning
• scalability
• inference cost

Innovation

• Mixture of Experts
• Continual Learning
• Channel State Information
• Scene-Adaptive
• Semantic Router