GNN-Based Prediction of Material Properties and Chemical Reactions

In this Merric webinar presentation, we will explore the application of Graph Neural Networks (GNN) in predicting material properties and chemical reactions. GNN has gained significant attention in recent years for its ability to model the characteristics and behavior of materials and molecules in the fields of chemistry and materials science. By effectively learning from graph-structured data, GNN finds widespread application in molecular structures and crystal structures, which can be easily represented as graphs. By incorporating both spatial and non-spatial information, GNN shows promising results in predicting properties such as band gap, formation energy, and reaction energy.

[@MERRICS]

55:53 말씀해주신 내용은 대부분 X에 대해서 y 값을 regression으로 찾는 것 같습니다.
혹시 역으로 원하는 y가 있을 때, design space 탐색한다는식의 최적의 x를 찾는 방식도 가능한가요?
원하는 output에 대한 다양한 후보 multimodal 혹은 Graph를 찾는 것도 재미있을 것 같아서요.
혹시 가능하시다면 그쪽으로는 어떤 연구가 진행되고 있는지? 찾아보는걸 추천하는지 여쭤보고 싶습니다!