[18p-A24-1~13] 23.1 Joint Session N "Informatics"

4:15 PM - 4:30 PM JST(7:15 AM - 7:30 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation Award

[18p-A24-10] Development of reliability evaluation for material image prediction using uncertainty

〇U OU¹, Hiroyuki Yamamoto¹, Sayaka Tanimoto¹ (1.Hitachi Ltd. Research & Development Group.)

Keywords:

Semantic Segmentation,uncertainty,materials informatics

In recent years, in the field of Materials Informatics which accelerates the development of new materials and devices, extracting features from structural images and calculating uncertainty are crucial tasks for constructing models to predict material properties. However, traditional uncertainty indicators were continuous and lacked clear targets for verification, leading to a heavy workload for predicting results. Therefore, we developed a method to automatically calculate the threshold of uncertainty in areas where prediction confidence is low. This greatly improved the efficiency of the verification process