Closing the Modality Gap in Zero-Shot HAR: Contrastive Training and Separability-Optimized Prototypes on IMU Data
arXiv:2606.10789v1 Announce Type: new Abstract: Zero-shot learning (ZSL) for inertial measurement unit (IMU)-based human activity recognition (HAR) faces a central challenge: bridging the gap between sensor embeddings and semantic class representations. We systematically evaluate seven configurations combining three inference methods with two training pipelines on the PAMAP2 dataset, using 14 seen and 4 unseen activity classes with subjects 108 and 109 held out for testing. We find that the modality gap is a training-time phenomenon governed by the encoder objective. A temporal convolutional n
The continuous advancements in AI and machine learning techniques are driving research into more efficient and robust zero-shot learning applications, particularly for real-world sensor data like IMUs.
Improving zero-shot learning for Human Activity Recognition (HAR) on IMU data can dramatically reduce the need for extensive labeled datasets, enabling faster deployment and broader application of AI in health monitoring, sports, and robotics.
This research provides a more systematic understanding of the 'modality gap' challenge in zero-shot HAR and offers practical solutions like contrastive training and separability-optimized prototypes to improve model performance.
- · AI researchers
- · Wearable technology companies
- · Healthcare sector
- · Sports analytics
Enhanced accuracy and efficiency for zero-shot human activity recognition using inertial measurement units.
Accelerated development and adoption of AI-powered applications in fields requiring real-time activity monitoring without extensive pre-training data.
Potentially democratized access to sophisticated activity recognition features for a wider range of devices and users, driving innovation in personalized health and performance tracking.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG