Abstract

In this talk, I will present a data-driven machine learning algorithm (Sun et al., JCTC , 18, 2022) that is devised to learn collective variables with a multitask neural network. In this work, new ways of labeling atomic configurations and approximating committor function are proposed. The resulting ML-learned collective variable is shown to be an effective low-dimensional representation, capturing the reaction progress and guiding effective umbrella sampling to obtain accurate free energy landscapes. This approach enables automated dimensionality reduction for energy controlled reactions in complex systems, offers a unified and data-efficient framework that can be trained with limited data, and outperforms single-task learning approaches, including autoencoders.