Room 505 Multifunction classroom, Cosmology Building, NTU
(臺灣大學次震宇宙館 505 多功能教室)
On the Vanishing Viscosity Approach to the Eikonal Equation via Neural Network
Hideo Kubo (Hokkaido University)
Abstract:
In this talk, I discuss numerical experiments on the Eikonal equation arising in mathematical
physics. Using Physics-Informed Neural Networks, we obtain accurate results for smoothly varying refractive indices, comparable to those of the Fast Marching Method.However, if the refractive index attains extremely large values in certain regions, the performance degrades.We show that introducing a viscosity term and tuning its parameter can substantially improve the results. This is a joint work with Prof. Koh Sato from Otaru University of Commerce.
Organizer: Chun-Hsiung Hsia (NTU)
