CaPriDe Learning

This repository is an implementation of CVPR 2023 paper titled: CaPriDe Learning: Confidential and Private Decentralized Learning based on Encryption-friendly Distillation Loss.

Dependencies

pip install -r requirements.txt

Run CaPriDe Learning

To train 5 models in CaPriDe learning protocol: (25 epochs correspond to the number of local training epochs.) Default dataset: CIFAR-10 (Homogeneous setting). To set the data setting to heterogeneous, simply change data_loader.py file, get_cifar10_train_loader() function.

python3 main.py --model_num 5 --is_train 1 --init_lr 0.1 --gamma 0.1 --use_gpu 1 --epochs 25 --resume 0 --save_name capride_cifar10_iid_p5_model 

Datasets

CIFAR-10 and CIFAR-100 datasets will be downloaded directly from torchvision.

Download HAM10000 dataset using this URL Link.

Encrypted Inference

To enable FHE scheme, refer to this link. To install it, you need to have Linux based Docker container (as a programming language you can choose either Python or C++).

Citation

@InProceedings{Tastan_2023_CVPR,
    author    = {Tastan, Nurbek and Nandakumar, Karthik},
    title     = {CaPriDe Learning: Confidential and Private Decentralized Learning Based on Encryption-Friendly Distillation Loss}, 
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
    pages     = {8084-8092}
}