用本地经验回放缓解联邦学习中的数据漂移遗忘问题
Federated Learning Under Temporal Drift -- Mitigating Catastrophic Forgetting via Experience Replay
- 客户端本地缓存历史样本，与当前数据混合训练
- 50样本/类缓冲区使准确率从28%回升至78-82%
- 无需修改服务器，适合长期更新的边缘设备场景
联邦学习在时间概念漂移下表现不佳，客户端数据分布随时间变化。我们发现，在时尚服饰数据集（Fashion-MNIST）上，标准的FedAvg方法在季节性漂移下会出现灾难性遗忘，准确率从74%下降至28%。为此，我们提出客户端侧的经验回放机制：每个客户端在本地训练时，将少量历史样本（缓冲区）与当前数据混合使用。该方法无需修改服务器聚合策略。实验表明，每类50个样本的缓冲区可使性能恢复至78%-82%，有效防止遗忘。消融研究显示，随着缓冲区增大，记忆与准确率之间存在明确权衡。
原文摘要 · Abstract (English)
Federated Learning struggles under temporal concept drift where client data distributions shift over time. We demonstrate that standard FedAvg suffers catastrophic forgetting under seasonal drift on Fashion-MNIST, with accuracy dropping from 74% to 28%. We propose client-side experience replay, where each client maintains a small buffer of past samples mixed with current data during local training. This simple approach requires no changes to server aggregation. Experiments show that a 50-sample-per-class buffer restores performance to 78-82%, effectively preventing forgetting. Our ablation study reveals a clear memory-accuracy trade-off as buffer size increases.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。