Transformer-Based Contrastive Meta-Learning For Low-Resource Generalizable Activity Recognition
Junyao Wang, Mohammad Abdullah Al Faruque
TL;DR
The paper tackles the challenge of generalizing HAR models under distribution shifts and limited labeled data. It introduces TACO, a transformer-based contrastive meta-learning framework that synthesizes virtual target domains within training batches, expands data diversity with sensor-time augmentations, and leverages a PatchTST encoder to extract expressive features. A supervised contrastive loss is embedded in the meta-learning objective, and the meta-optimization balances performance between meta-train and virtual target domains to improve cross-domain generalization. Empirical results on DSADS, PAMAP2, and USC-HAD show that TACO yields higher accuracy and greater robustness than state-of-the-art domain-generalization methods in low-resource scenarios, highlighting its practical impact for real-world HAR with privacy and labeling constraints.
Abstract
Deep learning has been widely adopted for human activity recognition (HAR) while generalizing a trained model across diverse users and scenarios remains challenging due to distribution shifts. The inherent low-resource challenge in HAR, i.e., collecting and labeling adequate human-involved data can be prohibitively costly, further raising the difficulty of tackling DS. We propose TACO, a novel transformer-based contrastive meta-learning approach for generalizable HAR. TACO addresses DS by synthesizing virtual target domains in training with explicit consideration of model generalizability. Additionally, we extract expressive feature with the attention mechanism of Transformer and incorporate the supervised contrastive loss function within our meta-optimization to enhance representation learning. Our evaluation demonstrates that TACO achieves notably better performance across various low-resource DS scenarios.