# Modeling probability distributions using an adversarial entropy system

Date:

04/17/2020 - 3:30pm to 5:00pm

Location:

Online

Speaker(s) / Presenter(s):

Ian Ellwood (Cornell)

Type of Event (for grouping events):

Title: Modeling probability distributions using an adversarial entropy system

Abstract: Producing a statistical model from a collection of samples drawn from an unknown probability distribution is one of the most common, yet still challenging problems in science. Recently, a new set of tools has emerged from machine learning which propose to solve this problem in cases that were considered intractable. One of the most interesting of these is the generative adversarial network or GAN. While GANs can train a generative network to produce high fidelity samples on datasets like photos, using them in science is currently obstructed by two problems: 1) the samples they generate have less entropy that the original distribution 2) It is difficult to reconstruct the probability distribution given the generator. In this seminar, I will discuss a new approach to the first problem and touch on aspects of the second.

Host: Das

Colloquium will be held by zoom. You will recieve link to talk via email.