Random sampling versus active learning algorithms for machine learning potentials of quantum liquid water

Add to your list(s) Download to your calendar using vCal

• Dr. Nore Stolte, Ruhr University Bochum
• Monday 03 February 2025, 14:30-15:00
• zoom.us/j/92447982065?pwd=RkhaYkM5VTZPZ3pYSHptUXlRSkppQT09.

If you have a question about this talk, please contact Eszter Varga-Umbrich.

Today, active learning is routinely used in training of machine learning potentials, but the efficacy of active learning across different systems is not well-tested. We have employed active learning algorithms based on committee disagreement for the training of a high-dimensional neural network potential for quantum liquid water at ambient conditions. Notably, we compare active to random data selection, given the same pool of candidate structures. I will discuss in detail the performance of active learning, at the level of computational requirements, train and test errors, and the quality of the final potentials in simulations including nuclear quantum effects.

This talk is part of the Lennard-Jones Centre series.

This talk is included in these lists:

• Hanchen DaDaDash
• Lennard-Jones Centre
• zoom.us/j/92447982065?pwd=RkhaYkM5VTZPZ3pYSHptUXlRSkppQT09

Note that ex-directory lists are not shown.