MuSDeT — Lightweight Wearable Stress Detection

Multi-scale encoding, gated fusion, and temporal context for stress detection on wearable biosignals. Published at MULA@CVPR 2026.

Venue: Multimodal Learning and Applications Workshop, CVPR 2026 (pp. 7454–7462). Role: First author. Paper: CVF Open Access · Code: Multimodal-Intelligence-Lab/MuSDeT · Poster: PDF

We propose a lightweight architecture for stress detection from wearable biosignals (EDA, HR, temperature, etc.) that combines multi-scale temporal encoding, gated cross-modal fusion, and a short-range temporal-context module. The design targets on-device inference on consumer wearables while maintaining competitive accuracy against heavier transformer-based baselines.