Property Prediction and Inverse Design of Polymers with HAPPY (Hierarchically Abstracted rePeat unit of PolYmers) Representation
발표자
()
초록
내용
Deep learning offers transformative insights into molecular structure-property relationships and advancing material design. However, its application to polymeric systems poses inherent challenges, due to complex interactions, vast combinatorial parameter spaces, and broad range of length and time scales. This presentation introduces an efficient polymer-to-string representation, HAPPY (Hierarchically Abstracted rePeat unit of PolYmers), which simplifies complex structures into manageable strings. HAPPY encapsulates essential polymer features by grouping sub-structures into single elements and linking them through grammatically complete and independent connectors. Integrated with Recurrent Neural Networks (RNN), HAPPY demonstrates robust and satisfactory performance in predicting bulk properties of polymer melts, even with limited datasets. Additionally, we explore HAPPY's potential in the inverse design of polymers, aiming to synthesize structures with specific target properties.