Juho Song

I am an M.S. student in the Graduate School of Data Science (GSDS) at KAIST, advised by Prof. Chanyoung Park. I received my B.S. in the Department of Mathematical Sciences from KAIST, with a minor in the School of Computing and the Entrepreneurship Program.

My research focuses on developing reliable and broadly applicable neural operator architectures for physics simulation, grounded in their mathematical foundations and geometric extensions. More broadly, I am interested in Scientific ML — including physics-informed surrogate modeling for complex systems, physically-constrained world models that learn to simulate physical dynamics from data, and inverse design where the goal is to find geometries or configurations that achieve target physical properties. My experience in the KAIST Entrepreneurship Program reinforces my drive to translate research into real-world engineering impact.

Recently, our paper EqGINO: Equivariant Geometry-Informed Fourier Neural Operators for 3D PDEs has been accepted at ICML 2026. [paper]

Research Interests

• Neural Operators for PDE Solving
• Physically-constrained World Models
• Inverse Problems & Design