Timezone: »

Wasserstein Variational Inference
Luca Ambrogioni · Umut Güçlü · Yağmur Güçlütürk · Max Hinne · Marcel A. J. van Gerven · Eric Maris

Thu Dec 06 07:45 AM -- 09:45 AM (PST) @ Room 210 #35

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 implicit 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.

Author Information

Luca Ambrogioni (Donders Institute)
Umut Güçlü (Donders Institute for Brain, Cognition and Behaviour, Radboud University)
Yağmur Güçlütürk (Donders Institute for Brain, Cognition and Behaviour, Radboud University)
Max Hinne (University of Amsterdam)
Marcel A. J. van Gerven (Radboud Universiteit)
Eric Maris (Donders Institute)

More from the Same Authors