Getting Started

The code runs on Python 3 with Tensorflow version 1 (>=1.13). To install the dependencies, run

pip3 install -r requirements.txt

Then, download the datasets manually and put them into the datasets folder.

• For MNIST dataset, download from http://yann.lecun.com/exdb/mnist/ and put the standalone files into datasets/mnist.
• For CIFAR-10 dataset, download the "CIFAR-10 binary version (suitable for C programs)" from https://www.cs.toronto.edu/~kriz/cifar.html, extract the standalone *.bin files and put them into datasets/cifar-10-batches-bin.
• For Fashion-MNIST dataset, download from https://github.com/zalandoresearch/fashion-mnist

To test the code:

• Run server.py and wait until you see Waiting for incoming connections... in the console output.
• Run some parallel instances of "client.py"/"fedprox_client.py"/"fedadm_client.py" on the same machine as the server.
• You will see console outputs on both the server and clients indicating message exchanges.

Code Structure

All configuration options are given in config.py which also explains the different setups that the code can run with.

The results are saved as CSV files in the results folder. The CSV files should be deleted before starting a new round of experiment. Otherwise, the new results will be appended to the existing file.

Currently, the supported datasets are MNIST(MNIST-O,MNIST-F) and CIFAR-10, and the supported models are SVM and CNN. The code can be extended to support other datasets and models too.

GPU Support

2 RTX 4090 and 6 RTX 2080 Ti GPUs were used for the experiments.

CUDA: 10.0 cuDNN: 7.4 python 3.7