Presentation Information
[8p-P11-6]Atomic Level Explanation via PFP Descriptors and Shapley Values
〇Bon Cho1 (1.Preferred Computational Chemistry, Inc.)
Keywords:
Materials Informatics,Explainable AI,Machine Learning Interatomic Potential
Materials Informatics is rapidly advancing materials discovery by leveraging machine learning. A primary challenge, however, is the "black-box" nature of many models, which hinders the derivation of new scientific knowledge due to their lack of explainability. In response, this work introduces an explainable approach that combines Shapley values with the PFP descriptors. These descriptors are atomic features derived from PFP, a universal machine learning interatomic potential.