Medical Imaging Benchmarks
Client data were partitioned with a Dirichlet distribution to simulate cross-silo heterogeneity. The model was a 4-layer CNN designed for 128 x 128 medical images and computational efficiency. Strong non-IID heterogeneity made unlearning h…
1 sources - 4 claims
Client data were partitioned with a Dirichlet distribution to simulate cross-silo heterogeneity. The model was a 4-layer CNN designed for 128 x 128 medical images and computational efficiency. Strong non-IID heterogeneity made unlearning harder but retained substantial utility. The experiments used OASIS, PathMNIST, and OrganAMNIST as medical imaging benchmarks.