Personalized Federated Learning towards Communication Efficiency, Robustness and Fairness

Authors: Shiyun Lin, Yuze Han, Xiang Li, Zhihua Zhang

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we conduct a large number of experiments to show the empirical superiority of our method over several state-of-the-art methods on the three aspects.
Researcher Affiliation Academia Shiyun Lin1,2 Yuze Han2 Xiang Li2 Zhihua Zhang1,2 1Center for Statistical Science, Peking University 2School of Mathematical Sciences, Peking University shiyunlin@stu.pku.edu.cn hanyuze97@pku.edu.cn lx10077@pku.edu.cn zhzhang@math.pku.edu.cn
Pseudocode Yes Algorithm 1 lp-proj: Projection-based Lp Regularized Personalized Federated Learning
Open Source Code Yes Source code for the reproduction of numerical results is available at https://github.com/desternylin/perfed.
Open Datasets Yes We test lp-proj as well as other comparable algorithms on six datasets from common ML and FL benchmarks [50, 8].
Dataset Splits Yes For each client, the training and testing data are pre-specified as in the ML community, and 20% of training data is randomly extracted to construct a validation set, keeping the remaining 80% as the training set.
Hardware Specification Yes All experiments are conducted on an NVIDIA RTX 3090 GPU.
Software Dependencies Yes The experiments are implemented with Python 3.8.13 and PyTorch 1.11.0.
Experiment Setup Yes More details about hyperparameter tuning are provided in Appendix C.2.