Methods with Local Steps and Random Reshuffling for Generally Smooth Non-Convex Federated Optimization

This repository provides the code to reproduce the results of the paper "Methods with Local Steps and Random Reshuffling for Generally Smooth Non-Convex Federated Optimization". It has three main scripts, which can be used to run 3 types of the experiments.

run_crr.py

This script can be used to reproduce results for the first set of experiments for Methods with random reshuffling subsection on $$f(x) = \frac{1}{N} \sum_{i=1}^{N} |x - x_i|^4,\ x_i \in [-10, 10]^d.$$

For information on arguments use python run_crr.py --help.

run_crr_torch.py

This script can be used to reproduce results for the second set of experiments for Methods with random reshuffling subsection on ResNet-18 on CIFAR10 dataset.

For information on arguments use python run_crr.py --help.

run_fl.py

This script can be used to reproduce results for the third and fourth sets of experiments for Methods with local steps and Methods with local steps, random reshuffling and partial participation on $$f(x) = \frac{1}{N} \sum_{i=1}^{N} |x - x_i|^4,\ x_i \in [-10, 10]^d.$$ All the clients in these experiments are spawned due to technical limitations.

For information on arguments use python run_fl.py --help.