FedRec

The source code and data for our paper "Untargeted Attack against Federated Recommendation Systems via Poisonous Item Embeddings and the Defense" in AAAI 2023.

Requirements

• PyTorch == 1.7.1
• pickle
• tqdm

Get Started

• Prepare Data

  You can prepare all the data by running the command bash prepare.sh under Data/. The script will download and preprocess the MovieLens-1M and Gowalla datasets for experiments.

• Run Experiments

  Code/run.sh is the script for running experiments.

  For training, you can run the command bash run.sh train under Code/. You can modify the value of hyper-parameters to change the setting of experiments. For example, by setting ATTACKER_STRAT as ClusterAttack, ATTACKER_RATIO as 0.01, and AGG_TYPE as NormBoundUNION, you can train the Federated Recommendation system with our MultiKrum+UNION defense mechanism against our ClusterAttack. Please refer to Code/train.py for more options.

  For testing, you can run bash run.sh test under Code/.