Date/Time
Date(s) - 04/03/2018
3:45 pm - 5:00 pm

Location
Rhines Hall, Room 125

Categories No Categories

Details

Join the Department of Materials Science & Engineering for light refreshments and a discussion lead by Dr. Bjørk Hammer of Aarhus University.

A number of search methods based on basin hopping or evolutionary algorithms are routinely used to identify, e.g. in a density functional theory framework, the most optimal cluster and surface structures for various inorganic compounds. In this talk, I introduce simple machine learning models and show how such models, when introduced in the search methodologies, do accelerate the finding of optimal structures. The machine learning models include unsupervised and supervised models, such as clustering[1], kernel enabled regression methods[2,3], and artificial neural networks[4,5]. Common the methods is a need for a proper representation of the compound structures and a discussion of different representations is hence taken, in particular with a view at the amount of data being available.