arxiveess.IVcs.CV2026-07-05
FedProIn: Mitigating Client Drift for Learnable Prototypes in Federated Medical Imaging
Harsh Kumar, Tarun Kumar Garg, Vaanathi Sundaresan
Federated learning (FL) is severely hindered by statistical heterogeneity due to variations in scanners, acquisition protocols, and patient populations. Such non-IID data induces client drift during local optimization, leading to unstable convergence and suboptimal global models…