Matthijs Vákár

Date: 16:00-17:00, Thursday, 03.03.2022

Location: MS Teams ICS Colloquium

Title:Automatic Differentiation

Abstract: Automatic differentiation (AD) tends to be the technique of choice whenever derivatives need to be computed in an efficient and numerically stable manner of a function implemented by a piece of code. Many applications in machine learning and scientific computing crucially depend on it. In recent years, programming language researchers such as the speaker have been studying AD to see how it can be simplified, applied to more expressive programming languages and how it can be proven correct. This talk will give a brief introduction to why AD is useful and how the technique works, and it will say a bit about the speaker’s work with Fernando Lucatelli Nunes and Tom Smeding on a simple, generally applicable and provably correct variant of AD that they have developed.