REVIEW 7 cited by
Towards Stability of Autoregressive Neural Operators
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
Neural operators have proven to be a promising approach for modeling spatiotemporal systems in the physical sciences. However, training these models for large systems can be quite challenging as they incur significant computational and memory expense -- these systems are often forced to rely on autoregressive time-stepping of the neural network to predict future temporal states. While this is effective in managing costs, it can lead to uncontrolled error growth over time and eventual instability. We analyze the sources of this autoregressive error growth using prototypical neural operator models for physical systems and explore ways to mitigate it. We introduce architectural and application-specific improvements that allow for careful control of instability-inducing operations within these models without inflating the compute/memory expense. We present results on several scientific systems that include Navier-Stokes fluid flow, rotating shallow water, and a high-resolution global weather forecasting system. We demonstrate that applying our design principles to neural operators leads to significantly lower errors for long-term forecasts as well as longer time horizons without qualitative signs of divergence compared to the original models for these systems. We open-source our \href{https://github.com/mikemccabe210/stabilizing_neural_operators}{code} for reproducibility.
Forward citations
Cited by 7 Pith papers
-
Explainable quantum-compressed machine learning for complex fluid flows
A hybrid quantum-classical surrogate compresses the latent time-stepping operator of flow models to as few as 8 trainable parameters and achieves stable long rollouts via exact unitarity, matching a classical baseline...
-
Autoregressive One-Step Generative Modeling for Dynamical System Forecasting
MeLISA extends pixel-space MeanFlow to one-step window-conditioned autoregressive forecasting, improving long-horizon turbulence statistics over neural-operator baselines.
-
MoWE : A Mixture of Weather Experts
MoWE, a ViT-based gating network, combines forecasts from Pangu, Aurora, and FCN3 with per-grid-point weights and beats each expert and the simple mean in RMSE.
-
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.
-
A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
3D-PRIMME learns a local grain-boundary update from two time steps on 100³ voxels and extrapolates to 1024³ domains while preserving coarsening kinetics and topology.
-
Diffeomorphic Neural Operator Learning
A neural operator that evolves fields by composing learned diffeomorphisms, enforcing relabeling symmetry and targeting conservative, non-diffusive turbulent forecasts.
-
Enhanced accuracy through ensembling of randomly initialized auto-regressive models for time-dependent PDEs
Deep ensembles of randomly initialized autoregressive models reduce long-horizon prediction error compared to any single model across three PDE-driven dynamical systems.
Discussion (0). Continue with ORCID to comment.