Generating nucleon-nucleus scattering data by Gaussian process regression

ガウス過程回帰を用いた核子-原子核散乱データの生成

渡辺 証斗*; 湊 太志   ; 木村 真明*; 岩本 信之  

Watanabe, Shoto*; Minato, Futoshi; Kimura, Masaaki*; Iwamoto, Nobuyuki

Recent progresses in the data science have greatly impacted the study of nuclear data evaluation. The AI-technologies have a possibility to improve the accuracy of nuclear data and reduce the human and time resources required to construct the database. As one of such challenges, we are building a machine learning system that optimizes and estimates parameters of the nucleon-nucleus scattering models to generate an AI-based nuclear database. In this contribution, we will explain how our system is designed and works effectively. Our system combines the Gaussian process regression with the CCONE code system. By fitting measured cross sections, it optimizes the parameters of the nuclear reaction models, such as the optical potential and structural parameters of a target nucleus. It also estimates unknown energy dependence of the model parameters from experimental data. We will demonstrate the performance of our system and also how it helps in creating nuclear databases.