FlashbackCL: Mitigating Temporal Forgetting in Federated Learning
2026-06-02 • Machine Learning
Machine LearningArtificial IntelligencePerformance
AI summaryⓘ
The authors study how to improve federated learning when the data on clients changes over time, which usual methods don’t handle well. They extend a recent method called Flashback by adding ways to forget older class counts gradually, store examples in a balanced way, and pick important examples on the server side. Their improved version, FlashbackCL, performs better on tests with changing data distributions and also helps when data doesn’t change. They find that the balanced replay buffer is the most important part of their improvements.
Federated LearningTemporal Distribution ShiftForgetting MitigationFlashbackClass-Balanced Reservoir SamplingReplay BufferCIFAR-10CIFAR-100Continual LearningActive Coreset Curation
Authors
Mubarak A. Ojewale, Adriana E. Chis, Jorge M. Cortes-Mendoza, Bernardo Pulido-Gaytan, Horacio Gonzalez-Velez
Abstract
Federated Learning (FL) of foundation and edge models increasingly targets deployments where client data distributions drift over time, yet existing forgetting-mitigation methods assume each client's distribution is stationary. Flashback, the strongest recent FL method against cross-client (spatial) forgetting, uses monotonically accumulating per-class label counts as a knowledge proxy; this proxy becomes miscalibrated under temporal distribution shift and anchors the global model to an outdated class balance. We formalise temporal forgetting in FL with a per-phase metric isolated from protocol-level fluctuations and propose Flashback Continual Learning (FlashbackCL), a drop-in extension of Flashback with (i) temporally-decayed label counts; (ii) a device-aware replay buffer with Class-Balanced Reservoir Sampling (CBRS); and (iii) server-side active coreset curation on the public distillation set. The results show that FlashbackCL achieves 6.9% to 10.0% relative improvement relative to Flashback, on CIFAR-10 with 50 clients and three controlled temporal shift modes, while simultaneously reducing temporal forgetting by up to 68%. A 5-variant ablation identifies CBRS replay as the critical component. FlashbackCL also improves Flashback by 3.5 points on stationary CIFAR-100, suggesting that class-balanced replay regularises spatial heterogeneity as well as temporal shift.