In recent years, Machine Learning (ML) algorithms have accurately reproduced energies derived
from quantum chemistry without the need to solve the Schrödinger equation. In this talk, I will
provide an overview of how these methods work and emphasize their speed and accuracy.
Examples from recent literature will be used to illustrate how ML can be used to perform reactive
molecular dynamics simulations on unprecedented length and time scales. Additionally, the concept
of "active learning" will be explored, which is where an ML algorithm is able to quantify it's own
accuracy and determine systematically improve itself with no human intervention. Then the ability
of ML algorithms to produce properties other than energies will be explored, such as atomic charges
and dipole moments. Finally, ongoing work where ML is used to generate effective Hamiltonian
parameters will be discussed.
Time: 10:00 - approx. 13:00
Location: University of Bremen, BCCMS, ECO5/TAB Building, Entrance F, Ground floor, room 0.50/0.51