Adversarial Tasks Paper: Optimal Transport

[ykobaya9]

Is your feature request related to a problem? Please describe. This issue stems from #428, where we are trying to explore techniques for progressive learner in dealing with adversarial tasks.

Describe the solution you'd like My idea was to explore optimal transport algorithms to align the data points or the posterior distribution. I would also like to compare the performance with the point cloud registration approaches I tried last semester. Lastly, my goal is to design a method for image datasets.

Describe alternatives you've considered

Additional context (e.g. screenshots) Some initial experiments show that optimal transport improves BTE for some degrees of rotation in R-XOR.

Wasserstein Distance:

Regularized Wasserstein Distance (Sinkhorn Knopp):