Jonathan Gallagher

Date: January 31, 2020

Time: 1:00 PM

Location: MS 427

Title: Taking the derivative of computations, backwards

Abstract:

This talk will introduce reverse differential restriction categories.  The reverse derivative is a fundamental operation in machine learning and differential programming.  Reverse differential categories provide an axiomatization of the reverse derivative.  In this talk, we will expand the axiomatic framework for reverse differentiation by combining it with restriction structure; this allows for reverse differentiation functions that may be partial (such as those defined by while-loops).