Welcome to the repository of the paper 'Generative Adversarial Learning of Sinkhorn Algorithm Initializations'.
The paper aims at warm-starting the Sinkhorn algorithm with initializations computed by a neural network, which is trained in an adversarial fashion similar to a GAN using a second, generating neural network. It is based on the Master's thesis 'A Sinkhorn-NN Hybrid Algorithm' by Jonathan Geuter, as well the follow up Master's Thesis 'Learning Optimal Transport Solutions with Deep neural networks' by Ingimar Tomasson.
In order reproduce the results of the paper it is as simple as executing the
experiment.py file. The required packages and their versions are detailed in the requirements.txt file.
The test data can be generated (scraped from online) by executing the make_data.py file or can be found at this Google Drive folder.
NB: This package is CUDA compatible