[2C04-06] Extraction Solvent for Minor Actinide Elements

[2C06] Exploration of Fluorinated Super-Solvents for Minor Actinide Extraction

(8)Accelerating Actinide Chemistry Research through Human-in-the-Loop Machine Learning

*Takahiro NISHIHARA¹, Masahiko Nakase¹, Takashi Kajitani¹, Tomohiro Okamura¹, Fauzia Hanum Ikhwan¹, Chihiro Tabata², Toru Kobayashi², Kota Matsui³, Hitomi Ishida⁴, Ryo Takahashi⁴ (1. Tokyo Tech, 2. JAEA, 3. Nagoya Univ., 4. MHI)

Keywords:Human In The Loop machine learnning, Transfer Learnning, Actinide Experiment

MA separation in next-generation reprocessing is important for cycle rationalization and reduction of the burden on the repository. We have aimed to accelerate actinide experiments by machine learning, in addition to searching for suitable solvents for MA extraction. Because even the acquisition of training data is difficult in some actinide experiments in Japan, utilizing historical data beyond past experimental conditions or deemed insufficient in performance is key. Inferring actinide experimental data from cold experiments or simulations could streamline research and implementing a “Human in the Loop” (HITL) approach with researcher insight is also crucial. We will show an example of its implementation that combines several machine learning schemes such as transfer learning to accelerate actinide chemistry research.