Seminar LAMBDA: Neural Networks for Structured Grid Generation: PINN and Weight Constraint approaches

Аннотация: 

Structured grids that parametrize the physical domain are essential for classes of PDE solvers that rely on tensor-product topology, such as finite-difference, spectral, and separation-of-variables methods.

In this seminar, we discuss the theory of structured grid generation for general non-convex domains, followed by AI-based implementations. Two approaches are presented: a PINN-based formulation and a weight-constrained neural network method. Both illustrate how neural networks can act as equation solvers and geometric parameterizations, rather than as standard feed-forward predictors trained only by physical loss.

Note. This seminar is not about solving physical PDEs on a mesh; instead, it focuses on constructing the mesh itself, which involves solving auxiliary equations.

Место проведения: АУК Покровский бульвар, 11, ауд. R408
Дата: 09.02.2026 
Время: 14:40-16:00