
Accelerating high-throughput simulations using machine learning methods
Thomas Bligaard
Recorded 29 May 2017 in Lausanne, Vaud, Switzerland
Course: High-throughput computations: general methods and applications using AiiDA - May, 2017
Abstract
Modern machine learning techniques allows one to fit (non-linear) models with many unknown parameters while maintaining control of issues relating to overfitting. When moving towards massively high-throughput simulations many parts of the simulations can benefit from being integrated with machine learning. Examples relate to improved initial guesses, faster search algorithms using surrogate machine learning models, systematic error estimation, reduction of model complexity, and improved simulation model accuracy.
Watched 1311 times.
Watch