/FedMM

Primary LanguagePython

This repository contains the code accompanying the paper " FedMM: Saddle Point Optimization for Federated Adversarial Domain Adaptation" Paperlink:

network structure

Requirements to run the code:

  1. Python 3.7
  2. Tensorflow 1.14.0
  3. numpy 1.20.3
  4. tqdm

Download dataset:

Download mnistm data:

curl -L -O http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/BSR/BSR_bsds500.tgz

Preprocess mnistm dataset

python create_mnistm.py

Experiments on Federated Domain Adaptation:

Usage for the Proposed FedMM on DANN loss:

python train.py -max_iter=15000 -lambda1_decay=1.05 -adv_loss='DANN'

Usage for the Proposed FedMM on MDD loss:

python train.py -max_iter=50000 -lambda1_decay=1.01 -adv_loss='MDD'

Usage for the Proposed FedMM on CDAN loss

python train.py -max_iter=30000 -lambda1_decay=1.02 -adv_loss='CDAN'

Reference

@misc{2110.08477,
Author = {Yan Shen, Jian Du, Han Zhao, Benyu Zhang, Zhanghexuan Ji, Mingchen Gao},
Title = {FedMM: Saddle Point Optimization for Federated Adversarial Domain Adaptation},
Year = {2021},
Eprint = {arXiv:2110.08477},
}