| |

This project is the code and the supplementary of "Federated Recommendation with Additive Personalization"

Precautions Before Use: FedRAP is highly sensitive to its hyperparameter combinations. Even slight deviations from the settings reported in the original paper can lead to substantial performance divergences. As such, practitioners should perform a fine-grained, dataset-specific hyperparameter search to reproduce the reported results and achieve optimal performance on their own benchmarks.

1. The code is implemented with Python >= 3.8 and torch~=1.13.1+cu117;
2. Other requirements can be installed by pip install -r requirements.txt.

1. First create two folders: ./logs and ./results;

2. Put datasets into the path [parent_folder]/datasets/;

3. python train.py --alias FedRAP --dataset movielens --data_file ml-100k.dat \
       --mu 1e-3 --l2_regularization 1e-6 --lr_network 1e-4 --lr_args 1e3
   

If you find this paper useful in your research, please consider citing:

@inproceedings{
    li2024federated,
    title={Federated Recommendation with Additive Personalization},
    author={Zhiwei Li and Guodong Long and Tianyi Zhou},
    booktitle={The Twelfth International Conference on Learning Representations},
    year={2024},
    url={https://openreview.net/forum?id=xkXdE81mOK}
}

• This project is free for academic usage. You can run it at your own risk.
• For any other purposes, please contact Mr. Zhiwei Li ()