Reducing Label Dependency in Human Activity Recognition with Wearables: From Supervised Learning to Novel Weakly Self-Supervised Approaches

The article explores methods to improve human activity recognition (HAR) using wearable devices by reducing the reliance on labeled data. It moves from traditional supervised learning to weakly self-supervised approaches, which is a significant area of research in AI, particularly in the context of sensor data and edge computing. The focus on weakly self-supervised learning suggests an attempt to improve model performance and reduce the cost of data annotation.

• •Focus on reducing label dependency in Human Activity Recognition.
• •Exploration of weakly self-supervised learning methods.
• •Application to wearable devices and sensor data.