Wasserstein Variational Inference [article]

Luca Ambrogioni, Umut Güçlü, Yağmur Güçlütürk, Max Hinne, Eric Maris, Marcel A. J. van Gerven
2018 arXiv   pre-print
This paper introduces Wasserstein variational inference, a new form of approximate Bayesian inference based on optimal transport theory. Wasserstein variational inference uses a new family of divergences that includes both f-divergences and the Wasserstein distance as special cases. The gradients of the Wasserstein variational loss are obtained by backpropagating through the Sinkhorn iterations. This technique results in a very stable likelihood-free training method that can be used with
more » ... t distributions and probabilistic programs. Using the Wasserstein variational inference framework, we introduce several new forms of autoencoders and test their robustness and performance against existing variational autoencoding techniques.
arXiv:1805.11284v2 fatcat:jqu4g4ldhvchfi2buqhscjnaly