Reproduce “Federated Continual Learning via Knowledge Fusion: A Survey” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xin Yang, Hao Yu, Xin Gao, Hao Wang, Junbo Zhang, Tianrui Li
This survey identifies and empirically validates 'spatial-temporal catastrophic forgetting' as a fundamental problem in Federated Continual Learning. It provides a structured overview of existing FCL methods, categorizing them by their knowledge fusion mechanisms to address this compounded forgetting challenge.
Key insights
- Spatial-temporal catastrophic forgetting emerges as a compounded challenge in Federated Continual Learning (FCL), combining temporal forgetting from sequential tasks and spatial forgetting from data heterogeneity across clients.
- Effective FCL solutions require sophisticated 'knowledge fusion' mechanisms beyond simple model aggregation, necessitating strategies like rehearsal, regularization, parameter isolation, or knowledge distillation.
- FCL frameworks can be broadly categorized into synchronous (clients share common task sequences) and asynchronous (clients have distinct, non-aligned task sequences), with asynchronous FCL being more practical and complex.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.