University of Michigan Researchers Open-Source ‘FedScale’: a Federated Learning (FL) Benchmarking Suite with Realistic Datasets and a Scalable Runtime to Enable Reproducible FL Research on Privacy-Pre

Overlooking just one component can cause the FL assessment to be skewed. Even though they have many datasets and FL training objectives (e.g., LEAF, their datasets frequently comprise synthetically created partitions derived from conventional datasets (e.g., CIFAR) and do not represent realistic features. FedScale Runtime to standardize and simplify FL assessment in more realistic conditions. FedScale Runtime is also extendable, allowing for the quick implementation of new algorithms and concepts through flexible APIs. Furthermore, it delivers datasets that properly simulate FL training scenarios where FL will be applied practically.