Distributional Variational AutoEncoder: Application to Synthetic Data Generation

This repository is the official implementation of Distributional Variational AutoEncoder: Application to Synthetic Data Generation (DistVAE) with pytorch.

NOTE: This repository supports WandB MLOps platform!

Training & Evaluation

1. How to Training & Evaluation

training

python main.py --dataset <dataset>

evaluation (step-by-step)

• Synthetic data generation and evaluation: synthesize.py
• CDF estimation and Vrate: inference.py
• Discretize estimated CDF example: calibration.py
• Membership inference attack for VAE:
  • shadow_data.py: generate training/test datasets for shadow models
  • shadow_main.py: training shadow models using training datasets from shadow_data.py
  • shadow_attack.py: evaluate membership inference attack for VAE

Directory and codes

.
+-- assets (for each dataset)
+-- modules 
|       +-- adult_datasets.py
|       +-- cabs_datasets.py
|       +-- covtype_datasets.py
|       +-- credit_datasets.py
|       +-- kings_datasets.py
|       +-- loan_datasets.py
|       +-- evaluation.py
|       +-- model.py
|       +-- simulation.py
|       +-- train.py

+-- main.py
+-- synthesize.py
+-- inference.py
+-- calibration.py
+-- shadow_data.py
+-- shadow_main.py
+-- shadow_attack.py
+-- LICENSE
+-- README.md

Citation

@inproceedings{
an2023distributional,
title={Distributional Learning of Variational AutoEncoder: Application to Synthetic Data Generation},
author={Seunghwan An and Jong-June Jeon},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=GxL6PrmEUw}
}