Search papers, labs, and topics across Lattice.
This paper tackles Federated Class-Incremental Learning (FCIL) in non-IID federated healthcare systems by introducing a dynamic memory allocation strategy for data replay. The strategy optimizes exemplar storage across clients based on data heterogeneity and performance fairness, aiming to mitigate catastrophic forgetting. Experiments on medical image datasets show that the proposed method, FeDMRA, achieves significant performance gains over existing FCIL baselines by rationally allocating limited storage resources.
Forget fixed memory budgets: dynamically allocating exemplar storage across federated clients boosts performance in class-incremental learning for heterogeneous medical data.
In federated healthcare systems, Federated Class-Incremental Learning (FCIL) has emerged as a key paradigm, enabling continuous adaptive model learning among distributed clients while safeguarding data privacy. However, in practical applications, data across agent nodes within the distributed framework often exhibits non-independent and identically distributed (non-IID) characteristics, rendering traditional continual learning methods inapplicable. To address these challenges, this paper covers more comprehensive incremental task scenarios and proposes a dynamic memory allocation strategy for exemplar storage based on the data replay mechanism. This strategy fully taps into the inherent potential of data heterogeneity, while taking into account the performance fairness of all participating clients, thereby establishing a balanced and adaptive solution to mitigate catastrophic forgetting. Unlike the fixed allocation of client exemplar memory, the proposed scheme emphasizes the rational allocation of limited storage resources among clients to improve model performance. Furthermore, extensive experiments are conducted on three medical image datasets, and the results demonstrate significant performance improvements compared to existing baseline models.
