[19p-C601-1~15] 23.1 Joint Session N "Informatics"

4:00 PM - 4:15 PM JST(7:00 AM - 7:15 AM UTC)

[19p-C601-10] Enhancing Inverse Problem Solutions with Accurate Surrogate Simulators and Promising Candidates

〇Akihiro Fujii¹, Hideki Tsunashima², Yoshihiro Fukuhara^2,3, Koji Shimizu¹, Satoshi Watanabe¹ (1.Tokyo Univ., 2.Waseda Univ., 3.ExaWizards Inc.)

Keywords:

inverse problem,artificial electromagnetic material,deep learning

Solving inverse problems using deep learning-based surrogate simulators (for the forward problem), we investigated the impact of surrogate simulator accuracy on the solutions of inverse problems. Then, we developed the NeuLag method, which can address the resource constraints of using high-accurate large-scale surrogate simulators. The NeuLag method efficiently finds optimal solutions using high-accurate large-scale simulators and reduces the resimulation error, the distance with the original data, by approximately 1/50 compared to conventional methods. Furthermore, we have examined the behavior under constraint conditions, demonstrating the potential to solve inverse problems with constraints.