This repository contains code for paper "On Minimax Optimality of GANs for Robust Mean Estimation" (AISTATS 2020).

We implemented f-GAN, MMD-GAN (with Gaussian kernel) and Wasserstein GAN (with Euclidean norm as ground cost). These models are tested under Huber's contamination model.

To install dependency, run

pip install -r requirements.txt

Run the following scripts containing detailed parameter configurations:

bash test_fgan.sh
bash test_mmd.sh
bash test_sinkhorn.sh