Contrastive training with rich text prototypes closes the modality gap in zero-shot IMU HAR, raising unseen accuracy from 58.3% to 73.2% and macro F1 from 0.34 to 0.583 on PAMAP2 with 4 held-out classes.
Limitations in employing natural language supervision for sensor-based human activity recognition— And ways to overcome them
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Closing the Modality Gap in Zero-Shot HAR: Contrastive Training and Separability-Optimized Prototypes on IMU Data
Contrastive training with rich text prototypes closes the modality gap in zero-shot IMU HAR, raising unseen accuracy from 58.3% to 73.2% and macro F1 from 0.34 to 0.583 on PAMAP2 with 4 held-out classes.