Automatic differentiation of Rust, with the friendly help of your compiler

(30 minute invited talk)

Manuel Drehwald (MIT Julia lab / University of Toronto)
Friday session 1 (Zoom) (13:0014:40 BST)
Watch a recording of this talk on YouTube

Calculating gradients is necessary for various algorithms such as back propagation, Bayesian inference, uncertainty quantification, and probabilistic programming. Enzyme is a tool that can automatically compute gradients and was accepted for experimental (nightly) integration into the Rust compiler. We demonstrate how rust-enzyme can be used, present the next development steps, and discuss Rust specific challenges for automatic differentiation.