User Tools

Site Tools


Statistics Seminar
Department of Mathematical Sciences

DATE:Thursday, Oct. 29, 2020
TIME:1:15pm – 2:15pm
LOCATION:Zoom meeting
SPEAKER:Zifan Huang, Binghamton University
TITLE:Categorical Reparameterization With Gumbel-Softmax


Categorical variables are a natural choice for representing discrete structure in the world. However, stochastic neural networks rarely use categorical latent variables due to the inability to backpropagate through samples. In this work, we present an efficient gradient estimator that replaces the non-differentiable sample from a categorical distribution with a differentiable sample from a novel Gumbel-Softmax distribution. This distribution has the essential property that it can be smoothly annealed into a categorical distribution. We show that our Gumbel-Softmax estimator outperforms state-of-the-art gradient estimators on structured output prediction and unsupervised generative modeling tasks with categorical latent variables, and enables large speedups on semi-supervised classification.

seminars/stat/201029.txt · Last modified: 2020/10/25 20:01 by qyu