This repository is the official implementation of TIFS 2022 paper: Towards Private Learning on Decentralized Graphs with Local Differential Privacy.
Wanyu Lin, Baochun Li, and Cong Wang. ”Towards Private Learning on Decentralized Graphs with Local Differential Privacy, ” in IEEE Transactions on Information Forensics & Security (TIFS), 2022.
This code is implemented in Python 3.9, and relies on the following packages:
If you want to individually train and evaluate the models on any of the datasets mentioned in the paper, run the following command:
python main.py -d cora -ee 8 -ex 1 -kx 4 -ky 2 --model sage --learning_rate 0.01 --weight_decay 0.001 --dropout 0.5 -s 1234 --max-epochs 500 -o ./out/cora/ee8_ex1
python main.py -d citeseer -ee 8 -ex 1 -kx 4 -ky 2 --model sage --learning_rate 0.01 --weight_decay 0.001 --dropout 0.5 -s 12345 --max-epochs 200 --orphic True -o ./out/citeseer/ee8_ex1
python main.py -d lastfm -ee 8 -ex 1 -kx 4 --model sage --learning_rate 0.01 --weight_decay 0.001 --dropout 0.5 -s 12345 --max-epochs 500 -o ./out/lastfm/ee8_ex1
dataset arguments:
-d <string> name of the dataset (choices: cora, pubmed, facebook, lastfm) (default: cora)
--data-dir <path> directory to store the dataset (default: ./datasets)
--data-range <float pair> min and max feature value (default: (0, 1))
--val-ratio <float> fraction of nodes used for validation (default: 0.25)
--test-ratio <float> fraction of nodes used for test (default: 0.25)
data transformation arguments:
-ex <float> privacy budget for feature perturbation (default: inf)
-ee <float> privacy budget for edge perturbation (default: inf)
model arguments:
--model <string> backbone GNN model (choices: gcn, sage, gat) (default: sage)
--hidden-dim <integer> dimension of the hidden layers (default: 16)
--dropout <float> dropout rate (between zero and one) (default: 0.0)
-kx <integer> KProp step parameter for features (default: 0)
-ky <integer> KProp step parameter for labels (default: 0)
--forward <boolean> applies forward loss correction (default: True)
trainer arguments:
--optimizer <string> optimization algorithm (choices: sgd, adam) (default: adam)
--max-epochs <integer> maximum number of training epochs (default: 500)
--learning-rate <float> learning rate (default: 0.01)
--weight-decay <float> weight decay (L2 penalty) (default: 0.0)
--patience <integer> early-stopping patience window size (default: 0)
--device <string> desired device for training (choices: cuda, cpu) (default: cuda)
--orphic <string> use estimator to clean graph (choices: True, False) (default: False)
--inner_epoch <integer> the time of using estimator in a single epoch (default: 2)
--checkpoint_dir <string> the path to save checkpoint
experiment arguments:
-s <integer> initial random seed (default: None)
-r <integer> number of times the experiment is repeated (default: 1)
-o <path> directory to store the results (default: ./output)
--log <boolean> enable wandb logging (default: False)
--log-mode <string> wandb logging mode (choices: individual,collective) (default: individual)
--project-name <string> wandb project name (default: Solitude)
--retrain <boolean> whether to retrain the model (default: False)
The test result for each run will be saved as a csv file in the directory specified by
-o option (default: ./output).
If you find this code useful, please cite the following paper:
@article{lin2022towards,
title={Towards private learning on decentralized graphs with local differential privacy},
author={Lin, Wanyu and Li, Baochun and Wang, Cong},
journal={IEEE Transactions on Information Forensics and Security},
volume={17},
pages={2936--2946},
year={2022},
publisher={IEEE}
}