Machine-learning of density functionals for applications in molecules and materials
Prof. Kieron Burke
Recorded 20 February 2018 in Lausanne, Vaud, Switzerland
Event: MARVEL NCCR - MARVEL Distinguished Lectures, Seminars and Tutorials - 2015 onwards
This lecture is designed to be accessible to a wide variety of backgrounds.
In the first part, I will briefly review density functional theory and why it is important to many branches of modern physical science. I will also review machine learning and its recent applications to molecules and materials.
In the second half, I will show how, in collaboration with computer scientists at TU Berlin, we have used a specific type of machine-learning, called kernel ridge regression, to find more accurate and powerful approximate density functionals than any made by humans.
For more background, see here.
Watched 1136 times.Watch