(169dc) Representation of Stochastic Polymeric Materials for Machine Learning Applications

Polymeric materials are versatile, with applications ranging from nano-science to everyday items. For industrial purposes, these materials often exhibit significant stochastic variations in molecular weight and composition. Representing such materials for machine learning purposes and computational studies remains a challenge. In this presentation, we explore a stochastic graph-based approach that can be constructed from BigSMILES and G-BigSMILES string representations. These graphs are then encoded via graph autoencoders to represent them in a continuous R^N space. This representation can serve as an initial point for further machine learning applications, such as deep neural nets, databank searches, or can be interpolated for new materials with generative AI.