I am a third-year student at the University of Toronto pursuing a specialist in computer science, a major in mathematics, and a minor in statistics. I am especially interested in Bayesian deep learning.
I have taken a fair amount of interesting as well as advanced courses across CS, math and stats. A comprehensive list can be found here.
A pdf version of my CV can be found here.
Discrete random variables are a natural choice to promote category selection in stochastic computation graphs. However, a major difficulty of adopting a stochastic node whose underlying random variable is discrete stems from the inability to efficiently backpropagate gradients through them during training. In this report, we survey the effectiveness of gradient estimation using the recently proposed Concrete/Gumbel-Softmax distribution.