Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Multi-Grid Tensorized Fourier Neural Operator for High-Resolution PDEs

About

Memory complexity and data scarcity have so far prohibited learning solution operators of partial differential equations (PDEs) at high resolutions. We address these limitations by introducing a new data efficient and highly parallelizable operator learning approach with reduced memory requirement and better generalization, called multi-grid tensorized neural operator (MG-TFNO). MG-TFNO scales to large resolutions by leveraging local and global structures of full-scale, real-world phenomena, through a decomposition of both the input domain and the operator's parameter space. Our contributions are threefold: i) we enable parallelization over input samples with a novel multi-grid-based domain decomposition, ii) we represent the parameters of the model in a high-order latent subspace of the Fourier domain, through a global tensor factorization, resulting in an extreme reduction in the number of parameters and improved generalization, and iii) we propose architectural improvements to the backbone FNO. Our approach can be used in any operator learning setting. We demonstrate superior performance on the turbulent Navier-Stokes equations where we achieve less than half the error with over 150x compression. The tensorization combined with the domain decomposition, yields over 150x reduction in the number of parameters and 7x reduction in the domain size without losses in accuracy, while slightly enabling parallelism.

Jean Kossaifi, Nikola Kovachki, Kamyar Azizzadenesheli, Anima Anandkumar• 2023

Related benchmarks

TaskDatasetResultRank
10-step incompressible flow rolloutNS-SL moderate viscosity (m) (test)
UV Relative L2 Ratio0.0101
30
10-step incompressible flow rolloutNS-SL extreme viscosity (x) (test)
UV relative-L2 ratio0.018
30
10-step incompressible flow rolloutNS-G extreme viscosity (x) (test)
UV Relative L2 Ratio0.0163
15
10-step incompressible flow rolloutNS-G moderate viscosity (m) (test)
UV relative-L2 ratio0.021
15
10-step incompressible flow rolloutNS-Sines moderate viscosity (m) (test)
UV Relative L2 Ratio0.489
15
10-step incompressible flow rolloutNS-Sines extreme viscosity (x) (test)
UV Relative L2 Ratio0.668
15
10-step incompressible flow rolloutNS-PwC moderate viscosity (m) (test)
UV Relative L2 Ratio1.113
15
10-step incompressible flow rolloutNS-PwC extreme viscosity (x) (test)
UV Relative L2 Ratio1.395
15
PDE System Rollout ForecastingActive Matter
VRMSE (01:16)4.443
14
10-step rollout predictionShear Flow
VRMSE0.89
14
Showing 10 of 13 rows

Other info

Follow for update