"Processed Results for Low-Visible Robust Federated Learning"
"This dataset contains the processed experimental results and figure-source files associated with the study \u201cLow-Visible Robust Weight Generation for Byzantine Federated Learning via Sketch-Based Detection.\u201d The data support the evaluation of CipherSketch-RFL under Byzantine model-update attacks, including sign flipping, scaling, model replacement, A Little Is Enough, and inner product manipulation. The package includes multi-seed accuracy summaries, Byzantine weight statistics, visibility-sensitivity results, protected-backend verification records, and source data for the manuscript figures. MNIST, Fashion-MNIST, and CIFAR-10 are public benchmark datasets and are not redistributed. They can be downloaded automatically through torchvision using the accompanying code package. The processed results are provided in CSV and JSON formats for inspection and reproduction."