Learnable Explicit Density for Continuous Latent Space and Variational Inference
Chin-Wei Huang Affiliation: MILA, Université de Montréal, Canada Correspondence to: Ahmed Touati Affiliation: MILA, Université de Montréal, Canada Laurent Dinh Affiliation: MILA, Université de Montréal, Canada Michal Drozdzal Affiliation: MILA, Université de Montréal, Canada Affiliation: Imagia Inc., Canada Mohammad Havaei Affiliation: MILA, Université de Montréal, Canada Affiliation: Imagia Inc., Canada Laurent Charlin Affiliation: MILA, Université de Montréal, Canada Affiliation: HEC Montréal, Canada Aaron Courville Affiliation: MILA, Université de Montréal, Canada Affiliation: CIFAR Fellow, Canada
Abstract
In this paper, we study two aspects of the variational autoencoder (VAE): the prior distribution over the latent variables and its corresponding posterior. First, we decompose the learning of VAEs into layerwise density estimation, and argue that having a flexible prior is beneficial to both sample generation and inference. Second, we analyze the family of inverse autoregressive flows (inverse AF) and show that with further improvement, inverse AF could be used as universal approximation to any complicated posterior. Our analysis results in a unified approach to parameterizing a VAE, without the need to restrict ourselves to use factorial Gaussians in the latent real space.
原文 arXiv:1710.02248;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1710.02248v1