Fuse-PINN: Combining Finite Element Methods and Physics-Informed Neural Networks to enable massively parallel simulations of non-linear PDEs
The authors report near-perfect multi-GPU scaling as problem size grows in tests of nonlinear physics relevant to fusion.
Fuse-PINN uses a finite-element formulation for neural networks to parallelise physics-informed simulations on GPU-based high-performance computing systems. Its demonstrations include nonlinear convective models with fine-scale dynamics relevant to magnetic confinement fusion. An Energy Natural Gradient Descent optimiser supports high-precision convergence. The reported scaling concerns runs with increasing problem size; the abstract does not quantify speedups.
ArXiv · AI/CL/LG's note
Fuse-PINN uses a finite-element formulation for neural networks to parallelise physics-informed simulations on GPU-based high-performance computing systems. Its demonstrations include nonlinear convective models with fine-scale dynamics relevant to magnetic confinement fusion. An Energy Natural Gradient Descent optimiser supports high-precision convergence. The reported scaling concerns runs with increasing problem size; the abstract does not quantify speedups.
ArXiv · AI/CL/LG's note
score 5