Abstract

Machine learning has emerged in recent years as a powerful tool for the description of complex chemical systems. Based on the well-established Behler-Parrinello neural network potential methodology for interatomic potentials, I show in this talk that it has multiple benefits to combine a number of machine learning models to a committee. Instead of a single model, multiple models yield an average that outperforms its individual members as well as a measure of the generalization error in the form of the committee disagreement.

This disagreement can not only be used to identify the most relevant configurations to build up the model’s training set in an active learning procedure, but can also be monitored and biased during simulations to control the generalization error.

This facilitates the adaptive development of committee neural network potentials and their training sets, while keeping the number of ab initio calculations to a minimum. In this talk I will show that this methodology enables the rapid development of robust and accurate machine learning potentials for complex aqueous systems.

• https://bham-ac-uk.zoom.us/j/85702415099?pwd=VTh5aFZ4Sm9Nc1dxZXYwelJpc1JtZz09
• Meeting ID: 857 0241 5099
• Passcode: 501623