FeDXL: Provable Federated Learning for Deep X-Risk Optimization

This is the official implementation of the paper "FeDXL: Provable Federated Learning for Deep X-Risk Optimization" published on ICML2023.

How to run

If you are using a cluster with SLURM scheduler:

sbatch run.slurm

otherwise, use

python -m torch.distributed.launch --nproc_per_node=4 --nnodes=4 --node_rank=0 --master_addr='YOUR IP' --master_port=8888 \
            main.py --T0=5000 --lr=0.1 --I=32 --total_iter=10000

Reference

This is an implementation of the following paper:

@article{guo2022fedx,
  title={FedX: Federated Learning for Compositional Pairwise Risk Optimization},
  author={Guo, Zhishuai and Jin, Rong and Luo, Jiebo and Yang, Tianbao},
  journal={arXiv preprint arXiv:2210.14396},
  year={2022}
}