Collaborative Multi-Task Learning across Internet Edges with Device-to-Device Communications
Ryusei Higuchi, Hiroshi Esaki, Hideya Ochiai · 2023
People make various reactions to a presented object. Some may talk about the object itself and others may talk about the color or size. Nowadays, such different kinds of reactions, coming from various cognitive axes, could be automatically associated with the presented object in a smart device and stored as personal data. Integration of multiple cognitive axes across smart devices allows the development of a brainstormer that can generate multiple reactions to a presented object at the same time. We propose collaborative multi-task learning that integrates multiple cognitive axes across the smart devices utilizing device-to-device communications inspired by wireless ad hoc federated learning (WAFL). We especially focus on weakly-labeled cases in a multi-task context, based on the fact that people provide ideas mainly from their major cognitive axes but not from minor ones. In our benchmark-based evaluation, MT-WAFL has shown a novel performance improving its accuracy by integrating external cognitive axes without centralized mechanisms.