Federated Meta-Learning: Methodologies and Directions
摘要
Federated Meta-Learning (FML), a fusion of Federated Learning (FL) and Meta-Learning principles, has emerged as a hot topic recently. It introduces a novel paradigm that makes global models personalized within fewer fine-tuning steps on the local dataset. This survey explores the domain of FML by analyzing the key motivations for FML and suggesting a unique taxonomy of FML techniques categorized according to their algorithms and applications. This paper highlights their key ideas and envisions promising future trajectories of research, specifically discussing Federated Meta Knowledge which is regarded as the object of study in FML.