/Confederated-Learning

Primary LanguagePython

Confederated Learning: Going Beyond Centralization

This is a Pytorch implementation of the methods described in our paper:

Z. Wang, Q. Xu, K. Ma, X. Cao and Q. Huang. Confederated Learning: Going Beyond Centralization. MM2022.

Dependencies

  • Pytorch >= 1.9.0
  • numpy

Data

We perform evaluations on the following benchmark datasets: (a)MNIST $\to$ USPS; (b) the Office-31 dataset; (c) Office $\to$ Caltech. For each dataset, we create a corresponding folder in the root folder, and the dataset.py describes how the data is organized.

For MNIST and USPS, you can get the data via Pytorch. For the Office-31 dataset and Office $\to$ Caltech, all the data can be found in their homepage.

Train and test

For each dataset, the source model is available via

python train_source.py

and its performance on the target dataset is available via

python test_source_on_target.py

Besides, you can train the baseline model and the proposed methods via

python train_target_raw.py
python train_target_reweighting.py
python train_target_ensemble.py
python train_target_reg.py

Citation

@inproceedings{DBLP:conf/mm/WangX0CH21,
  author    = {Zitai Wang and
               Qianqian Xu and
               Ke Ma and
               Xiaochun Cao and
               Qingming Huang},
  title     = {Confederated Learning: Going Beyond Centralization},
  booktitle = {{ACM} Multimedia Conference},
  year      = {2022},
}