ICLR Reproducibility Challenge: Generative Adversarial Models For Learning Private And Fair Representations
Branch: master
Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.ipynb_checkpoints cleaned up for publishing Jan 12, 2019
data cleaned up for publishing Jan 12, 2019
project_webpage cleaned up for publishing Jan 12, 2019
.gitignore
README.md cleaned up for publishing Jan 12, 2019
iclr_adult_reproduced.ipynb cleaned up for publishing Jan 12, 2019
iclr_har_reproduced.ipynb cleaned up for publishing Jan 12, 2019

README.md

ICLR Reproducibility Challenge: Generative Adversarial Models For Learning Private And Fair Representations

We present code reproducing some of the results for a paper submitted to ICLR'19 entitled Generative Adversarial Models for Learning Private and Fair Representations. This paper explores using a generative adversarial model as a decorrelation mechanism that hides a sensitive variable while still preserving the utility of the original data. In our work, we replicated the architecture described in the paper using PyTorch and show a successful use case for the Human Activity Recognition (HAR) dataset.

Repository Structure

The primary experiments and code are presented in an Jupyter Notebook, iclr_har_reproduced.ipynb. The notebook requires PyTorch, Pandas, numpy, and matplotlib to run.

This project was completed in part as a final project for the Georgia Tech class CS7643 Deep Learning. As part of our submission, we created a website explaining our process and results.

We also attempted to apply the model to a novel dataset, the UCI Adult dataset, with limited results in iclr_har_reproduced.ipynb.

Researchers

Name Affiliation
Angel Alexander Cabrera Georgia Tech
Varun Gupta Georgia Tech
Will Epperson Georgia Tech

Links