/DProxSGT

Primary LanguagePython

DProxSGT

This repository is the implementations of the paper: Compressed Decentralized Proximal Stochastic Gradient Method for Nonconvex Composite Problems with Heterogeneous Data (Yan et al. 2023)

  • RegL1_FMNIST: code for training a sparse neural network model (LeNet5) on Fashion-MNIST with L1-regularlization using non-compressed methods

  • RegL1_CIFAR10: code for training a sparse neural network model (FixupResnet20) on CIFAR10 with L1-regularlization using non-compressed methods

  • Compress_FMNIST: code for training a neural network model (LeNet5) on Fashion-MNIST using compressed methods

  • Compress_CIFAR10: code for training a neural network model (FixupResnet20) on CIFAR10 using compressed methods

Running the file run.sh in each directory will produce the results. The figures in the figures file are generated based on these results.

Reference