针对医疗图像联邦学习，提出个性化单次通信训练新方法。
FedBiCross: Personalized One-Shot Federated Learning on Medical Images
- 按模型输出相似性聚类客户端，构建一致子群体
- 双层跨集群优化，自适应融合有益知识并抑制负迁移
- 支持个性化蒸馏，适合非独立同分布数据场景
基于无数据知识蒸馏的单次通信联邦学习（OSFL）可在不共享原始数据的前提下完成单轮训练，适用于隐私敏感的医疗应用。然而，现有方法将所有客户端预测聚合为全局教师，在非独立同分布（non-IID）数据下，冲突预测相互削弱，导致软标签信息量降低，弱化蒸馏效果。本文提出FedBiCross，一个三阶段个性化OSFL框架：(1) 根据模型输出相似性聚类客户端，形成一致子集成；(2) 双层跨集群优化，学习自适应权重，选择性利用有益的跨集群知识，同时抑制负迁移；(3) 客户端个性化蒸馏，实现特定适配。在四个医学图像数据集上的实验表明，FedBiCross在不同非独立同分布程度下均持续优于现有最优基线。
原文摘要 · Abstract (English)
Data-free knowledge distillation-based one-shot federated learning (OSFL) trains a model in a single communication round without sharing raw data, making OSFL attractive for privacy-sensitive medical applications. However, existing methods aggregate predictions from all clients to form a global teacher. Under non-IID data, conflicting predictions dilute each other during averaging, yielding less informative soft labels that weaken distillation. We propose FedBiCross, a personalized OSFL framework with three stages: (1) clustering clients by model output similarity to form coherent sub-ensembles, (2) bi-level cross-cluster optimization that learns adaptive weights to selectively leverage beneficial cross-cluster knowledge while suppressing negative transfer, and (3) personalized distillation for client-specific adaptation. Experiments on four medical image datasets demonstrate that FedBiCross consistently outperforms state-of-the-art baselines across different non-IID degrees.
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。