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