Federated Unlearning with Oriented Saliency Compression

🔗 paper link: Federated Unlearning with Oriented Saliency Compression (IJCNN 2025)

---

Prerequisites

Before running the code, ensure you have the following libraries installed:

pip install torch torchvision numpy pyyaml

File Struction

FedUOSC/
├── config/
│   ├── [dataset.yaml]      # configuration files
├── code/
│   ├── [model.py]          # Model definitions
│   ├── [utils.py]          # Utility functions
│   ├── [unlearning.py]     # Unlearning methods
│   ├── [main.py]           # Main script
├── [README.md]             

How to Use

1. Configuration File: prepare a .yaml configuration file before running the program or you can use the default file in ./config, where cu means client removal and su means sample removal in federated system.

2. Run the Program: execute the following command to start the program:

   python code/main.py --config <path_to_config_file> --setting <configuration_name>
   

   for example:

   python code/main.py --config config/cifar10.yaml --setting cu_iid
   

Notes

• Datasets will be automatically downloaded to the specified path (change your dataset path in ./code/utils.py).
• Adjust the parameters in the configuration file as needed.