Embedding-reparameterization procedure for manifold-valued latent variables in generative models

Dec 06, 2018

Eugene Golikov, Maksim Kretov

Conventional prior for Variational Auto-Encoder (VAE) is a Gaussian distribution. Recent works demonstrated that choice of prior distribution affects learning capacity of VAE models. We propose a general technique (embedding-reparameterization procedure, or ER) for introducing arbitrary manifold-valued variables in VAE model. We compare our technique with a conventional VAE on a toy benchmark problem. This is work in progress.
Dec 06, 2018

Eugene Golikov, Maksim Kretov

**Click to Read Paper**

Differentiable lower bound for expected BLEU score

Aug 23, 2018

Vlad Zhukov, Eugene Golikov, Maksim Kretov

Aug 23, 2018

Vlad Zhukov, Eugene Golikov, Maksim Kretov

**Click to Read Paper**

Using stochastic computation graphs formalism for optimization of sequence-to-sequence model

Dec 15, 2017

Eugene Golikov, Vlad Zhukov, Maksim Kretov

Dec 15, 2017

Eugene Golikov, Vlad Zhukov, Maksim Kretov

**Click to Read Paper**