CoSy lunch seminar 11/02/2020

Abstract: Atomic neural networks (ANNs) constitute a class of machine learning methods for predicting potential energy surfaces and physio-chemical properties of molecules and materials.  Recently, we have taken the initiative and developed an open-source Python library named PiNN (https://github.com/Teoroo-CMC/PiNN/), allowing researchers to easily develop and train state-of-the-art ANN architectures for making chemical predictions. In particular, we designed and implemented an interpretable and high-performing graph convolutional neural network architecture PiNet, and demonstrated how the chemical insight “learned” by such a network can be extracted (Shao Y., Hellström M., Mitev P. D., Knijff L., Zhang C., 2020, J. Chem. Inf. Model., DOI: 10.1021/acs.jcim.9b00994).