Symbolic Regression for Discovery of a DFT Functional

Patrick Riley
Relay Therapeutics

Symbolic regression is a family of machine learning algorithms that aim to produce small mathematical expressions that fit observed data.
In this talk, I will review the SyFES system we recently published that brings together several ideas from work in AutoML and applies them to finding a new DFT functional.
We show that we can rediscover the B97 exchange functional, then show an improvement on the popular wB97M-V functional.
The audience does not need to be interested in DFT to appreciate the demonstration of machine guided search for a compact model in a scientifically important domain.

Presentation (PDF File)

Back to Learning and Emergence in Molecular Systems