Simplified Polymer-to-String Representation for Properties Prediction Deep Learning Network
발표자
()
초록
내용
Deep learning has gained traction as a tool to elucidate the relationship between the structure and property of materials. However, in contrast to their inorganic counterparts, organic polymer systems possess highly complex chemical/topological structures and multi-level features. This intricacy in the structure made applying deep learning to polymer fundamentally challenging. In this study, we propose a condensed and intuitive polymer-to-string representation and demonstrate its efficiency and robustness in constructing property prediction RNN(Recurrent Neural Network). Our result indicates that implementing deep learning with the proposed representation generates satisfactory performance in identifying important features relevant to predicting dielectric constant, solubility, glass transition temperature, thermal conductivity, and density of polymer melts.