adsgan

Codebase for Anonymization through Data Synthesis using Generative Adversarial Networks (ADS-GAN)

Authors: Jinsung Yoon, Lydia N. Drumright, Mihaela van der Schaar

Reference: Jinsung Yoon, Lydia N. Drumright, Mihaela van der Schaar, "Anonymization through Data Synthesis using Generative Adversarial Networks (ADS-GAN): A harmonizing advancement for AI in medicine," IEEE Journal of Biomedical and Health Informatics (JBHI), 2019.

Paper Link: https://ieeexplore.ieee.org/document/9034117

Contact: jsyoon0823@gmail.com

This directory contains implementations of ADSGAN framework for synthetic data generation.

To run the pipeline for training and evaluation on ADSGAN framwork, simply run python3 -m main_adsgan.py.

Code explanation

(1) data_loader.py

• Load and preprocess original data

(2) adsgan.py

• Generate synthetic data using the original data

(3) feature_distribution.py

• Compare feature distribution between original data and synthetic data

(4) compute_wd.py

• Compare Wasserstein distance between original data and synthetic data

(5) compute_identifiability.py

• Compare Identifiability between original data and synthetic data

(6) main_adsgan.py

• Report the performances of ADS-GAN in terms of Wasserstein Distance, Identifiability, and Distribution.

Command inputs:

• iterations: Number of experiments iterations
• lamda: Hyper-parameter to control the identifiability and quality of the synthetic data
• h_dim: Number of hidden state dimensions
• z_dim: Number of random state dimensions
• mb_size: Number of mini-batch samples

Note that hyper-parameters should be optimized for different datasets.

Example command

$ python3 main_adsgan.py --iterations 10000 --lamda 0.1 --h_dim 30
--z_dim 10 --mb_size 128

Outputs

• orig_data: original data
• synth_data: synthetically generated data
• measures: performances of 3 metrics (Distribution, WD, identifiability)