Flashback: Understanding and Mitigating Forgetting in Federated Learning
This addresses a critical issue in Federated Learning for applications with heterogeneous client data, representing a strong incremental improvement.
The paper tackles the problem of forgetting in Federated Learning, which hampers convergence under data heterogeneity, and proposes Flashback, an algorithm that mitigates forgetting and achieves faster convergence, requiring only 6 to 16 rounds to reach target accuracy.
In Federated Learning (FL), forgetting, or the loss of knowledge across rounds, hampers algorithm convergence, particularly in the presence of severe data heterogeneity among clients. This study explores the nuances of this issue, emphasizing the critical role of forgetting in FL's inefficient learning within heterogeneous data contexts. Knowledge loss occurs in both client-local updates and server-side aggregation steps; addressing one without the other fails to mitigate forgetting. We introduce a metric to measure forgetting granularly, ensuring distinct recognition amid new knowledge acquisition. Leveraging these insights, we propose Flashback, an FL algorithm with a dynamic distillation approach that is used to regularize the local models, and effectively aggregate their knowledge. Across different benchmarks, Flashback outperforms other methods, mitigates forgetting, and achieves faster round-to-target-accuracy, by converging in 6 to 16 rounds.