Try to improve the performance of FedMD
Based on Tensorflow 2.0 and Python 3.7
Use various json files to specify the type of training to be performed.
Save Emnist dataset.
Pretrain models with public data and their own private data.
According to the corresponding json file to run federated learning.
Includes various federal learning models, such as FedMD and its variants.
-
The origin implementation of FedMD training.
-
Select stochastic batches of models to calculate consensus.
-
Use cosine similarity function to calculate consensus.
-
Each client only learns on its private dataset.
Run the pre_train for instance
python pre_train.py -conf conf/pre_train.json