Clustering-based Mutual-Learning for Personalized Situation-Aware Services in Smart Homes

The Internet of Things (IoT) has been extensively applied to human-centric smart environments. Services provisioned within these IoT-enabled smart settings can substantially enhance the quality of life, mitigate potential hazards, and thereby offer personalized services for their users. However, there is a notable deficiency in the consideration of human factors necessary for realizing more refined and personalized situation-aware services. Moreover, as learning-based approaches are widely used in providing situation analysis in the current era, the challenge of training a robust learning model is aggravated by the scarcity of locally collected user data. Federated Learning (FL) was proposed to address the issue in a centralized, cloud-edge-based setting. Nonetheless, it falls short of facilitating personalized learning, which is crucial for the provisioning of local situation-aware services. In this paper, we propose a decen-tralized, clustering-based mutual learning approach that enables each edge server to learn a personalized model by iteratively sharing knowledge within clusters formed based on situational similarities. We used connected smart homes as an example to demonstrate the learning approach, and show the effectiveness of achieving personalized situation analysis, which ultimately leads to robust and accurate service in smart environments.