4 ms·
Physics-informed neural networks (PINNs) [video]
- peter_d_sherman 3y agoPaper: "Physics Informed Deep Learning (Part I): Data-driven Solutions of Nonlinear Partial Differential Equations" (by Maziar Raissi, Paris Perdikaris, and George Em Karniadakis) https://arxiv.org/abs/1711.10561 https://arxiv.org/abs/1711.10561 https://arxiv.org/pdf/1711.10561.pdf https://arxiv.org/pdf/1711.10561.pdf >"The resulting neural networks form a new class of data-efficient universal function approximators that naturally encode any underlying physical laws as prior information." Also: >"In this first part, we demonstrate how these networks can be used to infer solutions to partial differential equations" This is some seriously, seriously impressive work!