사물인터넷 환경에서 효과적인 분할 연합 학습을 위한 통신 오버헤드 최적화 연구 동향

초록

Split federated learning has attracted significant attention in Internet of Things (IoT) environments, as it enables resource-constrained devices to participate in training without sharing raw data. By utilizing data from IoT devices, intelligent services can be provided in healthcare, manufacturing, cities, and transportation. However, split federated learning incurs communication overhead due to the repeated exchange of training-related information during the training process. This overhead increases training time and energy consumption in IoT environments with limited bandwidth and energy resources, becoming a major constraint on practical deployment. Thus, recent studies on optimizing the communication overhead of split federated learning have been conducted. This paper introduces major communication overhead problems in split federated learning for IoT environments and systematically compares solution strategies for the specific problems addressed in each study.

키워드