University of Cambridge > Talks.cam > Theory - Chemistry Research Interest Group > Rapid development of robust single purpose machine learning potentials for complex aqueous systems

Rapid development of robust single purpose machine learning potentials for complex aqueous systems

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

If you have a question about this talk, please contact Lisa Masters.

While machine learning has opened various new avenues in computational chemistry and material science, it usually still remains a difficult task to obtain robust and accurate models for a given system of interest. Here, we show that committee neural network potentials can solve this problem by providing readily developed single purpose models for complex molecular systems. Using active learning based on query by committee techniques, a new model can be obtained in a one-step process from a single reference trajectory. We apply this methodology based on committee models to multiple aqueous systems with increasing complexity. The six aqueous systems chosen here comprise different ions in solution, water in nanotubes and on titanium dioxide interfaces, as well as water under MoS2 confinement. Highlighting the accuracy of our approach, the resulting models are validated in detail with a new scoring scheme that includes structural and dynamical properties and the precision of the force prediction of the models. By making the underlying packages available, we think that such single purpose committee models will enable the uncomplicated but accurate extension of time and length scales in molecular simulations.

This talk is part of the Theory - Chemistry Research Interest Group series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2021 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity