Expected to improve the practicality and scalability of wearable devices

Proposes a zero-shot human activity recognition (Zero-Shot HAR) AI technology based on wearable sensors

Lim Gyu-yeon, a third-year Computer Engineering major at Gachon University, recently published a paper as the first author in the top-tier SCI international journal IEEE Sensors Journal (IF 4.5). The corresponding author is Professor Lee Myung-kyu from Hanyang University’s Biomedical Engineering Research Institute.

The paper, titled “Contrastive Learning for Zero-Shot Human Activity Recognition Using Labeled Simple Actions on Wearable Devices,” proposes a zero-shot learning (ZSL) AI model for human activity recognition using sensor data collected from wearable devices.

The study addresses the limitations of existing zero-shot learning (ZSL) approaches, which rely on external semantic information such as attribute annotations or textual descriptions. By employing contrastive learning, the research proposes a method to learn a structured embedding space directly from raw sensor signals. This approach captures temporal and structural similarities, enabling the system to recognize complex activities or perform new zero-shot classifications that were not observed during training—without requiring explicit semantic information—demonstrating the potential for a scalable human activity recognition system.

This research is expected to significantly enhance the practicality and scalability of wearable devices. By overcoming the limitation of conventional models that are restricted to pre-defined activity classes, it lays the groundwork for flexible adaptation to new activity patterns in fields such as healthcare and rehabilitation monitoring.

Paper link: https://ieeexplore.ieee.org/document/11184457