Tensor-Train Compressed Separable PINNs: A Curvature-Aware Optimization Framework for Parametric PDEs in High Dimensions
My rating
The resource targets high-dimensional parametric partial differential equations using physics-informed neural networks. Its tensor-train compressed separable framework is designed around coordinate-separable neural architectures and linear differential operators. The method introduces a Gauss-Newton pullback metric as an operator-informed curvature measure in parameter space and relates the resulting optimization scheme to natural-gradient approaches.
Tags: