guanzhongliu/fedmd_imp

Try to improve the performance of FedMD

FedMD_Imp

Try to improve the performance of FedMD

Based on Tensorflow 2.0 and Python 3.7

Introduction

conf(Directory)

Use various json files to specify the type of training to be performed.

dataset(Directory)

Save Emnist dataset.

pretrain.py

Pretrain models with public data and their own private data.

model_train.py

According to the corresponding json file to run federated learning.

FedMD.py

Includes various federal learning models, such as FedMD and its variants.

• FedMD

  The origin implementation of FedMD training.

• FedMD_random

  Select stochastic batches of models to calculate consensus.

• FedMD_simu

  Use cosine similarity function to calculate consensus.

• FedMD_own

  Each client only learns on its private dataset.

How to run

Run the pre_train for instance

python pre_train.py -conf conf/pre_train.json

fedmd_imp's Issues

ValueError: Layer model expects 1 input(s), but it received 2 input tensors.