Materials Expert-Artificial Intelligence for Materials Discovery

• DOI: 10.48550/arxiv.2312.02796
• Publication Date: 2023-01-01
• Authors: Y. Liu; Milena Jovanović; Krishnanand Mallayya; Wesley J. Maddox; Andrew Gordon Wilson; Sebastian Klemenz; Leslie M. Schoop; Eun-Ah Kim
• Keywords: Intuition

The advent of material databases provides an unprecedented opportunity to uncover predictive descriptors for emergent properties from vast data space. However, common reliance on high-throughput ab initio necessarily inherits limitations such data: mismatch with experiments. On the other hand, experimental decisions are often guided by expert's intuition honed experiences that rarely articulated. We propose using machine learning "bottle" operational into quantifiable expertly curated measurement-based data. introduce "Materials Expert-Artificial Intelligence" (ME-AI) encapsulate and articulate this human intuition. As a first step towards program, we focus topological semimetal (TSM) among square-net materials as property inspired expert-identified descriptor based structural information: tolerance factor. start curating dataset encompassing 12 primary features 879 materials, whenever possible. then use Dirichlet-based Gaussian process regression specialized kernel reveal composite semimetals. ME-AI learned independently reproduce expert expand upon it. Specifically, new point hypervalency critical chemical feature predicting TSM within compounds. Our success carefully defined problem points "machine bottling insight" approach promising learning-aided discovery.