FedPerf: A Practitioners’ Guide to Performance of Federated Learning Algorithms
NeurIPS 2020 Workshop on Pre-registration in Machine Learning, PMLR 148:302-324, 2021.
Federated Learning (FL) enables edge devices to collaboratively train a global model without sharing their local data. This decentralized and distributed approach improves user privacy, security, and trust. Different variants of FL algorithms have presented promising results on both IID and skewed Non-IID data. However, the performance of FL algorithms is found to be sensitive to the FL system parameters and hyperparameters of the used model. In practice, tuning the right set of parameter settings for an FL algorithm is an expensive task. In this preregister paper, we propose an empirical investigation on four prominent FL algorithms to discover the relation between the FL System Parameters (FLSPs) and their performances. The FLSPs add extra complexity to FL algorithms over a traditional ML system. We hypothesize that choosing the best FL algorithm for the given FLSP is not a trivial problem. Further, we endeavor to formulate a systematic method that could aid the practitioners in selecting a suitable algorithm given the FLSPs. The code for all the experiments is available here: https://github.com/tushar-semwal/fedperf.