Authored by Fritsch, S. et al. | Published by Ulster University | 2025

This research introduces a hierarchical zero-shot learning method for Human Activity Recognition (HAR) in smart homes. It addresses privacy, sustainability, and adaptability concerns by using lightweight language models rather than cloud-based systems. The approach is designed to be efficient and transferable, supporting applications like health monitoring and independent living. It aims to reduce reliance on high-energy cloud platforms while enabling accurate, privacy-conscious activity detection in residential environments.

Keywords: