REVIEW 11 cited by
Towards Multi-spatiotemporal-scale Generalized PDE Modeling
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Partial differential equations (PDEs) are central to describing complex physical system simulations. Their expensive solution techniques have led to an increased interest in deep neural network based surrogates. However, the practical utility of training such surrogates is contingent on their ability to model complex multi-scale spatio-temporal phenomena. Various neural network architectures have been proposed to target such phenomena, most notably Fourier Neural Operators (FNOs), which give a natural handle over local & global spatial information via parameterization of different Fourier modes, and U-Nets which treat local and global information via downsampling and upsampling paths. However, generalizing across different equation parameters or time-scales still remains a challenge. In this work, we make a comprehensive comparison between various FNO, ResNet, and U-Net like approaches to fluid mechanics problems in both vorticity-stream and velocity function form. For U-Nets, we transfer recent architectural improvements from computer vision, most notably from object segmentation and generative modeling. We further analyze the design considerations for using FNO layers to improve performance of U-Net architectures without major degradation of computational cost. Finally, we show promising results on generalization to different PDE parameters and time-scales with a single surrogate model. Source code for our PyTorch benchmark framework is available at https://github.com/microsoft/pdearena.
Forward citations
Cited by 11 Pith papers
-
Neptuna: A Comprehensive Machine Learning Framework for Benchmarking Complex Multiphase Flows
A new 2.4 TB benchmark shows no single ML surrogate dominates on shock-driven multiphase flows, and composite losses with SoftAdapt weighting improve interface and spectral fidelity.
-
FluxNet: Learning Capacity-Constrained Local Transport Operators for Conservative and Bounded PDE Surrogates
FluxNet is a family of neural PDE time-steppers whose learned transport plans make discrete conservation exact and one-sided bound preservation structural, with empirical dual-bound control via a consistency-regulariz...
-
Bubbleformer: Forecasting Boiling with Transformers
Bubbleformer is a spatiotemporal transformer that autonomously forecasts boiling fields across fluids and regimes, and its authors release the BubbleML 2.0 dataset to support further work.
-
PDESpectralRefiner: Achieving More Accurate Long Rollouts with Spectral Adjustment
PDESpectralRefiner applies blurring diffusion with a new velocity prediction formula to refine PDE rollout outputs, yielding slightly lower rollout error than the prior diffusion refiner on 2D Navier-Stokes.
-
Hierarchical Implicit Neural Emulators
Feeding a hierarchy of predicted coarse-grained future states into an autoregressive neural emulator greatly improves long-term stability for 2D turbulent flow forecasting.
-
Hierarchical-embedding autoencoder with a predictor (HEAP) as efficient architecture for learning long-term evolution of complex multi-scale physical systems
HEAP, a hierarchical autoencoder with a predictor that advances multiple scale layers in sync, achieves several-fold lower long-term rollout error than flat ResNet baselines on Hasegawa-Wakatani turbulence.
-
Governing Equation Discovery from Data Based on Differential Invariants
PDE discovery guided by symmetry can be done by building the SINDy library from the differential invariants of the PDE's symmetry group, which shrinks the search space and improves success rates.
-
LLT: Local Linear Transformer for PDE Operator Learning
Local Linear Transformer learns PDE operators by combining linear global attention with local spatial mixing, achieving competitive accuracy and lower training-step cost than prior transformers.
-
Conditional Clifford-Steerable CNNs for PDE Modeling
Conditional Clifford-Steerable CNNs, which condition the equivariant kernel on pooled input features, improve PDE forecasting accuracy but do not prove the claimed complete kernel basis.
-
SlotPi: Physics-informed Object-centric Reasoning Models
SlotPi combines a learned Hamiltonian energy module with spatiotemporal attention to improve object-centric video prediction and visual question answering on several datasets.
-
Principled Approaches for Extending Neural Architectures to Function Spaces for Operator Learning
A practical recipe to convert common neural architectures into discretization-agnostic neural operators, validated by Navier-Stokes experiments showing cross-resolution generalization of FNO-style models.
Discussion (0). Sign in to comment.