Conceptual diagram of the proposed machine learning-based framework for alloy discovery (IMAGE)
Caption
This study presents a hybrid framework that integrates materials data with AI-extracted scientific knowledge, enabling uncertainty-aware discovery. Evidence about elemental substitutions in alloys is collected from two independent sources: material datasets, where alloy pairs with matching properties indicate substitutable elements, and large language models queried across five scientific domains. These “evidence streams” are combined using Dempster–Shafer theory to evaluate candidate alloys while explicitly quantifying prediction confidence versus uncertainty, guiding researchers toward promising candidates while flagging regions where current knowledge is insufficient.
Credit
Hieu-Chi Dam from Japan Advanced Institute of Science and Technology, Japan.
Usage Restrictions
Cannot be reused without permission.
License
Original content