Federated Composite Optimization

This code repository is for "Federated Composite Optimization" authored by Honglin Yuan (Stanford), Manzil Zaheer (Google) and Sashank Reddi (Google), to appear in ICML 2021.

talk video at FLOW seminar | slides (pdf)

bibtex

@inproceedings{DBLP:conf/icml/YuanZR21,
  author    = {Honglin Yuan and
               Manzil Zaheer and
               Sashank J. Reddi},
  editor    = {Marina Meila and
               Tong Zhang},
  title     = {Federated Composite Optimization},
  booktitle = {Proceedings of the 38th International Conference on Machine Learning,
               {ICML} 2021, 18-24 July 2021, Virtual Event},
  series    = {Proceedings of Machine Learning Research},
  volume    = {139},
  pages     = {12253--12266},
  publisher = {{PMLR}},
  year      = {2021},
  url       = {http://proceedings.mlr.press/v139/yuan21d.html},
  timestamp = {Wed, 14 Jul 2021 15:41:58 +0200},
  biburl    = {https://dblp.org/rec/conf/icml/YuanZR21.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

---

This repo is forked from Federated Research repository developed with TensorFlow Federated (TFF), an open-source framework for machine learning and other computations on decentralized data.

Some pip packages are required by this library, and may need to be installed:

pip install absl-py
pip install attr
pip install dm-tree
pip install numpy
pip install pandas
pip install tensorflow
pip install tensorflow-federated

We also require Bazel in order to run the code. Please see the guide here for installation instructions.

Directory structure

The directory utils is forked from Federated Research repository, which contains general utilities used by the other directories under research/. Examples include utilities for saving checkpoints, and configuring experiments via command-line flags.

The directory optimization is broken up into five task directories. Each task directory contains task-specific libraries (such as libraries for loading the correct dataset), as well as libraries for performing federated and non-federated (centralized) training. These are in the optimization/{task} folders.

A single binary for running these tasks can be found at main/federated_trainer.py. This binary will, according to absl flags, run any of the six task-specific federated training libraries.

The optimization/shared directory with utilities specific to these experiments, such as implementations of metrics used for evaluation.