Search NASA⌕ Search

SEARCH · Search NASA

Results for “deep ensemble”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

Uncertainty guided online ensemble for non-stationary data streams in fusion science

Machine Learning (ML) is poised to play a pivotal role in the development and operation of next-generation fusion devices. Fusion data shows non-stationary behavior with distribution drifts, resulted by both experimental evolution and machine wear-and-tear. ML models assume stationary distribution and fail to maintain performance when encountered with such non-stationary data streams. Online learning techniques have been leveraged in other domains, however it has been largely unexplored for fusion applications. In this paper, we investigate online learning for continuous adaptation to drifting data streams in the prediction of Toroidal Field (TF) coils deflection at the DIII-D fusion facility. We further address the short-term performance degradation inherent to standard online learning, which arises because ground truth is unavailable at prediction time. To mitigate this issue, we propose an uncertainty-guided online ensemble framework. The method leverages the Deep Gaussian Process Approximation (DGPA) for calibrated uncertainty estimation and uses these uncertainty measures to guide a meta-algorithm that aggregates predictions from learners trained over different historical horizons. Our results show that online learning reduces prediction error by 80% compared to a static model. The online ensemble and the proposed uncertainty-guided ensemble further reduce error by approximately 6%, and 10% respectively, relative to standard single-model online learning, while also providing calibrated uncertainty estimates to support operational decision-making.

AI↗
DOE OSTI2025

Uncertainty based Online Ensemble on Non-Stationary Data for Fusion Science

Machine Learning (ML) is poised to play a pivotal role in the development and operation of next-generation fusion devices. Fusion data shows non-stationary behavior due to drifts in the data. The drifts can arise from both experimental evolution and machine wear-and-tear. ML models assume stationary distribution and fail to maintain performance when encountered with non-stationary data streams.Online learning can be used to continuously adapt the models with new data as it is acquired. However, traditional online learning can suffer from short-term performance degradation, as ground truth are not available before making the prediction. To address this challenge, we propose uncertainty aware ensemble approach for online learning. We use Deep Gaussian Process Approximation (DGPA) technique for calibrated uncertainty estimation and use the uncertainty values to guide a meta-algorithm that produces predictions based on ensemble of learners. Moreover, DGPA also provides uncertainty estimation along with the predictions for decision makers. This paper demonstrates that the proposed method outperforms traditional online learning approach, and a naive ensemble without uncertainty guidance by about 7% and 6%, respectively, on B-coil deflection prediction at DIII-D Fusion Facility.

Rajput, Kishansingh [Thomas Jefferson National Acc↗
DOE OSTI2025

Uncertainty based Online Ensemble on Non-Stationary Data for Fusion Science

Machine Learning (ML) is poised to play a pivotal role in the development and operation of next-generation fusion devices. Fusion data shows non-stationary behavior due to drifts in the data. The drifts can arise from both experimental evolution and machine wear-and-tear. ML models assume stationary distribution and fail to maintain performance when encountered with non-stationary data streams.Online learning can be used to continuously adapt the models with new data as it is acquired. However, traditional online learning can suffer from short-term performance degradation, as ground truth are not available before making the prediction. To address this challenge, we propose uncertainty aware ensemble approach for online learning. We use Deep Gaussian Process Approximation (DGPA) technique for calibrated uncertainty estimation and use the uncertainty values to guide a meta-algorithm that produces predictions based on ensemble of learners. Moreover, DGPA also provides uncertainty estimation along with the predictions for decision makers. This paper demonstrates that the proposed method outperforms traditional online learning approach, and a naive ensemble without uncertainty guidance by about 7% and 6%, respectively, on B-coil deflection prediction at DIII-D Fusion Facility.

Rajput, Kishansingh [Thomas Jefferson National Acc↗
DOE OSTI2025

Resolving Mesoscale Convective Systems: Grid Spacing Sensitivity in the Tropics and Midlatitudes

Abstract Mesoscale convective systems (MCSs) are a critical global water cycle component and drive extreme precipitation events in tropical and midlatitude regions. However, simulating deep convection remains challenging for modern numerical weather and climate models due to the complex interactions of processes from microscales to synoptic scales. Recent models with kilometer‐scale horizontal grid spacings offer notable improvements in simulating deep convection compared to coarser‐resolution models. Still, deficiencies in representing key physical processes, such as entrainment, lead to systematic biases. Additionally, evaluating model outputs using process‐oriented observational data remain difficult. This study presents an ensemble of MCS simulations with spanning the deep convective gray zone ( from 12 km to 125 m) in the Southern Great Plains of the U.S. and the Amazon Basin. Comparing these simulations with Atmospheric Radiation Measurement (ARM) wind profiler observations, we find greater sensitivity in the Amazon Basin compared to the Great Plains. Convective drafts converge structurally at sub‐kilometer scales, but some deficiencies remain. In both regions, simulated up and downdrafts are too deep and extreme downdrafts are not strong enough. Furthermore, Amazonian updrafts are too strong. Overall, we observe higher sensitivity in the tropics, including an artificial buildup in vertical kinetic energy at scales of , suggesting a need for 250 m in this region. Nevertheless, bulk convergence—agreement of storm‐average statistics—is achievable with kilometer‐scale simulations within a 10% error margin with 1 km providing a good balance between accuracy and computational cost.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2024

Maximum Entropy Principle in Deep Thermalization and in Hilbert-Space Ergodicity

We report universal statistical properties displayed by ensembles of pure states that naturally emerge in quantum many-body systems. Specifically, two classes of state ensembles are considered: those formed by (i) the temporal trajectory of a quantum state under unitary evolution or (ii) the quantum states of small subsystems obtained by partial, local projective measurements performed on their complements. These cases, respectively, exemplify the phenomena of “Hilbert-space ergodicity” and “deep thermalization.” In both cases, the resultant ensembles are defined by a simple principle: The distributions of pure states have maximum entropy, subject to constraints such as energy conservation, and effective constraints imposed by thermalization. We present and numerically verify quantifiable signatures of this principle by deriving explicit formulas for all statistical moments of the ensembles, proving the necessary and sufficient conditions for such universality under widely accepted assumptions, and describing their measurable consequences in experiments. We further discuss information-theoretic implications of the universality: Our ensembles have maximal information content while being maximally difficult to interrogate, establishing that generic quantum state ensembles that occur in nature hide (scramble) information as strongly as possible. Our results generalize the notions of Hilbert-space ergodicity to time-independent Hamiltonian dynamics and deep thermalization from infinite to finite effective temperature. Our work presents new perspectives to characterize and understand universal behaviors of quantum dynamics using statistical and information-theoretic tools.

Eigenstate thermalization↗
DOE OSTI2025

Joint state-parameter estimation for the reduced fracture model via the united filter

Here, in this paper, we introduce an effective United Filter method for jointly estimating the solution state and physical parameters in flow and transport problems within fractured porous media. Fluid flow and transport in fractured porous media are critical in subsurface hydrology, geophysics, and reservoir geomechanics. Reduced fracture models, which represent fractures as lower-dimensional interfaces, enable efficient multi-scale simulations. However, reduced fracture models also face accuracy challenges due to modeling errors and uncertainties in physical parameters such as permeability and fracture geometry. To address these challenges, we propose a United Filter method, which integrates the Ensemble Score Filter (EnSF) for state estimation with the Direct Filter for parameter estimation. EnSF, based on a score-based diffusion model framework, produces ensemble representations of the state distribution without deep learning. Meanwhile, the Direct Filter, a recursive Bayesian inference method, estimates parameters directly from state observations. The United Filter combines these methods iteratively: EnSF estimates are used to refine parameter values, which are then fed back to improve state estimation. Numerical experiments demonstrate that the United Filter method surpasses the state-of-the-art Augmented Ensemble Kalman Filter, delivering more accurate state and parameter estimation for reduced fracture models. This framework also provides a robust and efficient solution for PDE-constrained inverse problems with uncertainties and sparse observations.

Bayesian inference↗
NASA NTRS1984

Using GPS and VLBI technology to maintain 14 digit synchronization

To facilitate the navigation of spacecraft to the outer planets, Jupiter and beyond, the JPL-NASA Deep Space Network (DSN) has implemented three ensembles of atomic clocks at widely separated locations. These clocks must be maintained, synchronized, to with a few parts in 10 to the 13th power of each other and, the entire group must be maintained, to a lesser degree, in synchronism with Coordinated Universal Time (UTC)NBS/USNO. Over the last 1 1/2 years the DSN has been using Global Positioning Satellites (GPS) and Very Long Baseline Interferometry (VLBI) technology to perform these critical Frequency and Time (F&T) synchronization tasks. A year of F&T synchronization data collected from the intercomparison of 3 sets of cesium and hydrogen maser driven clock ensembles through the use of GPS and VLBI techniques are covered. Also covered, are some of the problems met and limitations of these two techniques at their present level of technology.

Ward, S. C.↗
DOE OSTI2025

LTAU-FF: Loss Trajectory Analysis for Uncertainty in atomistic Force Fields

Model ensembles are effective tools for estimating prediction uncertainty in deep learning atomistic force fields. However, their widespread adoption is hindered by high computational costs and overconfident error estimates. In this work, we address these challenges by leveraging distributions of per-sample errors obtained during training and employing a distance-based similarity search in the model latent space. Our method, which we call LTAU (Loss Trajectory Analysis for Uncertainty), efficiently estimates the full probability distribution function of errors for any test point using the logged training errors, achieving speeds that are 2–3 orders of magnitudes faster than typical ensemble methods and allowing it to be used for tasks where training or evaluating multiple models would be infeasible. We apply LTAU towards estimating parametric uncertainty in atomistic force fields (LTAU-FF), demonstrating that it produces well-calibrated confidence intervals and predicts errors that correlate strongly with the true errors for data near the training domain. Furthermore, we show that the errors predicted by LTAU-FF can be used in practical applications for detecting out-of-domain data, tuning model performance, and predicting failure during simulations. We believe that LTAU will be a valuable tool for uncertainty quantification in atomistic force fields and is a promising method that should be further explored in other domains of machine learning.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Comparison of CNN-Based Image Classification Approaches for Implementation of Low-Cost Multispectral Arcing Detection

Camera-based sensing has benefited in recent years from developments in machine learning data processing methods, as well as improved data collection options such as Unmanned Aerial Vehicles (UAV) mounted sensors. However, cost considerations, both for the initial purchase of sensors as well as updates, maintenance, or potential replacement if damaged, can limit adoption of more expensive sensing options for some applications. To evaluate more affordable options with less expensive, more available, and more easily replaceable hardware, we examine the use of machine learning-based image classification with custom datasets, utilizing deep learning based-image classification and the use of ensemble models for sensor fusion. Utilizing the same models for each camera to reduce technical overhead, we showed that for a very representative training dataset, camera-based detection can be successful for detection of electrical arcing. We also use multiple validation datasets, based on conditions expected to be of varying difficulty, to evaluate custom data. These results show that ensemble models of different data sources can mitigate risks from gaps in training data, though the system will be less redundant for those cases unless other precautions are taken. We found that with good quality custom datasets, data fusion models can be utilized without specialization in design to the specific cameras utilized, allowing for less specialized, more accessible equipment to be utilized as multispectral camera components. This approach can provide an alternative to expensive sensing equipment for applications in which lower-cost or more easily replaceable sensing equipment is desirable.

convolutional neural networks↗
DOE OSTI2024

Evidential Deep Learning for Probabilistic Modelling of Extreme Storm Events

Uncertainty quantification (UQ) methods play an important role in reducing errors in weather forecasting. Conventional approaches in UQ for weather forecasting rely on generating an ensemble of forecasts from physics-based simulations to estimate the uncertainty. However, it is computationally expensive to generate many forecasts to predict real-time extreme weather events. Evidential Deep Learning (EDL) is an uncertainty-aware deep learning approach designed to provide confidence about its predictions using only one forecast. It treats learning as an evidence acquisition process where more evidence is interpreted as increased predictive confidence. We apply EDL to storm forecasting using real-world weather datasets and compare its performance with traditional methods. Our findings indicate that EDL not only reduces computational overhead but also enhances predictive uncertainty. This method opens up novel opportunities in research areas such as climate risk assessment, where quantifying the uncertainty about future climate is crucial.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Analyzing and Exploring Training Recipes for Large-Scale Transformer-Based Weather Prediction

Abstract The rapid rise of deep learning (DL) in numerical weather prediction (NWP) has led to a proliferation of models which forecast atmospheric variables with comparable or superior skill than traditional physics-based NWP. However, among these leading DL models, there is a wide variance in both the training settings and architecture used. Further, the lack of thorough ablation studies makes it hard to discern which components are most critical to success. In this work, we show that it is possible to attain high forecast skill even with relatively off-the-shelf architectures, simple training procedures, and moderate compute budgets. Specifically, we train a minimally modified Swin Transformer V2 (SwinV2) on ERA5 data and find that it attains superior skill in terms of mean-square errors of deterministic forecasts when compared against the European Centre for Medium-Range Weather Forecasts’ Integrated Forecasting System (IFS). Almost all DL–NWP systems share a core set of hyperparameters and design decisions. To aid and expedite future DL–NWP research, we present an in-depth, systematic exploration of different loss functions, model sizes and depths, patch sizes, and multistep training objectives. We also examine the model performance with metrics beyond the typical accuracy (ACC) and RMSE and investigate how the performance scales with model size. Through our open-source code, scoring pipelines, and models, we share our findings on key aspects of the training pipeline. These ablations reduce the necessity for expensive hyperparameter tuning and lower the barrier to entry for future DL–NWP research. Significance Statement This study investigates the potential of using large-scale transformer-based models for weather prediction, showing that it is possible to achieve high forecast accuracy with simpler, off-the-shelf architectures. By training a minimally modified SwinV2 transformer on ERA5 data, we show that the model achieves competitive forecast skill in terms of mean-square error for key variables, outperforming the European Centre for Medium-Range Weather Forecasts’ Integrated Forecasting System (IFS) at all lead times. Our findings suggest that effective training strategies, such as multistep fine-tuning and channel-weighted losses, significantly enhance the model’s performance. However, we also highlight that these improvements come with trade-offs in other areas, such as ensemble spread and high-frequency spatial detail. This work highlights the promise of deep learning in improving weather forecasts, which could lead to better preparedness and response to weather events, ultimately benefiting society by providing more reliable weather predictions.

Willard, Jared D. [Lawrence Berkeley National Labo↗
DOE OSTI2025

Karhunen–Loève deep learning method for surrogate modeling and approximate Bayesian parameter estimation

We evaluate the performance of the Karhunen-Loève Deep Neural Network (KL-DNN) framework for surrogate modeling and approximate Bayesian parameter estimation in partial differential equation models. In the surrogate model, the Karhunen-Loève (KL) expansions are used for the dimensionality reduction of the number of unknown parameters and variables, and a deep neural network is employed to relate the reduced space of parameters to that of the state variables. The KL-DNN surrogate model is used to formulate a maximum-a-posteriori-like least-squares problem, which is randomized to draw samples of the posterior distribution of the parameters. We test the proposed framework for a hypothetical unconfined aquifer via comparison with the forward MODFLOW and inverse PEST++ iterative ensemble smoother (IES) solutions as well as the state-of-the-art Fourier neural operator (FNO) and deep operator networks (DeepONets) operator learning surrogate models. Our results show that the KL-DNN surrogate model outperforms FNO and DeepONet for forward predictions. For solving inverse problems, the randomized algorithm provides the same or more accurate Bayesian predictions of the parameters than IES as evidenced by the higher log-predictive probability of both the estimated parameter field and the forecast hydraulic head. The posterior mean obtained from the randomized algorithm is closer to the reference parameter field than that obtained with FNO as the maximum a posteriori estimate.

Approximate Bayesian inference↗
DOE OSTI2025

The Importance of an Ensemble Approach for Modeling Aerosol‐Convection Interactions

Aerosol-convection interactions modulate cloud microphysics, thermodynamics, and updraft intensity, contributing to climate-scale aerosol-radiative forcing. However, quantifying aerosol indirect effects in mixed-phase deep convection remains challenging due to uncertainties in parameterized physics and initial conditions driving nonlinear evolution of convective processes. This study investigates the convective updraft sensitivity to random initial temperature perturbations using an idealized ensemble modeling framework informed by in situ thermodynamic and aerosol observations from the DOE TRACER field campaign. We analyze the impact of small-scale initial perturbations on updraft velocity and supersaturation, and determine the ensemble size required to minimize stochastic internal variability. Results show that minor thermodynamic perturbations can produce updraft variability comparable to aerosol-induced changes reported in prior work. An ensemble of 10 members sufficiently reduces variability, enabling robust investigation of aerosol-related updraft invigoration and informing the design of future ensemble-based aerosol-convection interaction studies in environments with significant spatiotemporal mesoscale thermodynamic and aerosol heterogeneity.

aerosol-convection interactions↗
DOE OSTI2026

The remarkable inefficiency of stratocumulus

Marine stratocumulus clouds play a central role in Earth's climate system by reflecting incoming solar radiation and exerting a strong cooling effect. Their organization into open and closed mesoscale cellular morphologies can be thought of as an example of bistable dynamics driven by aerosol–cloud interactions and mesoscale processes. From the perspective of non-equilibrium thermodynamics, these structures are an example of a far-from-equilibrium open system that continuously produces and exports entropy. While entropy production has been studied in idealized deep convective systems, it has not yet been quantified for shallow clouds. Here, we compute and decompose the internal entropy production of open- and closed-cell stratocumulus using an ensemble of large-eddy simulations. We show that the overall entropy production of stratocumulus is low, reflecting the limited vertical extent and corresponding reduced ability to utilize the energy fluxes at the system's boundaries. Moist processes dominate the overall irreversibility, which, combined with their low entropy production, leads to a mechanical efficiency about an order of magnitude smaller than in deep convective systems. Although the dominant irreversible processes differ between open- and closed-cell regimes, the distributions of total entropy production largely overlap across the ensemble, limiting the ability to distinguish the dynamics of individual cases based solely on total entropy production.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2024

How well are hazards associated with derechos reproduced in regional climate simulations?

Abstract. A 15-member ensemble of convection-permitting regional simulations of the fast-moving and destructive derecho of 29–30 June 2012 that impacted the northeastern urban corridor of the USA is presented. This event generated 1100 reports of damaging winds, generated significant wind gusts over an extensive area of up to 500 000 km2, caused several fatalities, and resulted in widespread loss of electrical power. Extreme events such as this are increasingly being used within pseudo-global-warming experiments to examine the sensitivity of historical, societally important events to global climate non-stationarity and how they may evolve as a result of changing thermodynamic and dynamic contexts. As such it is important to examine the fidelity with which such events are described in hindcast experiments. The regional simulations presented herein are performed using the Weather Research and Forecasting (WRF) model. The resulting ensemble is used to explore simulation fidelity relative to observations for wind gust magnitudes, spatial scales of convection (as is manifest in high composite reflectivity, cREF), and both rainfall and hail production as a function of model configuration (microphysics parameterization, lateral boundary conditions (LBCs), start date, use of nudging, compiler choice, damping, and number of vertical levels). We also examine the degree to which each ensemble member differs with respect to key mesoscale drivers of convective systems (e.g., convective available potential energy and vertical wind shear) and critical manifestations of deep convection, e.g., vertical velocities, cold-pool generation, and how those properties relate to the correct characterization of the associated atmospheric hazards (wind gusts and hail). Use of a double-moment, seven-class scheme with number concentrations for all species (including hail and graupel) results in the greatest fidelity of model-simulated wind gusts and convective structure to the observations of this event. All ensemble members, however, fail to capture the intensity of the event in terms of the spatial extent of convection and the production of high near-surface wind gusts. We further show very high sensitivity to the LBCs employed and specifically that simulation fidelity is higher for simulations nested within ERA-Interim compared to ERA5. Excess convective available potential energy (CAPE) in all ensemble members after the derecho passage leads to excess production of convective cells, wind gusts, cREF > 40 dBZ, and precipitation during a frontal passage on the subsequent day. This event proved very challenging to forecast in real time and to reproduce in the 15-member hindcast simulation ensemble presented here. Future work could examine if simulations with other initial and lateral boundary conditions can achieve greater fidelity.

Shepherd, Tristan (ORCID:0000000186276419)↗
DOE OSTI2025

Convection-Permitting Ensembles of an Isolated Mountain Thunderstorm during RELAMPAGO/CACTI

Abstract The north–south-oriented Sierras de Córdoba (SDC) ridge in central Argentina is noted for initiating thunderstorms that may grow into intense mesoscale convective systems (MCSs). It also initiates more isolated, shorter-lived cells under weaker synoptic forcing. These cells are less impactful than MCSs but may be difficult to predict in convective-scale numerical weather prediction (NWP) due to their strong sensitivities to subgrid and partially resolved processes. To study the mechanisms and predictability of such cells, convection-permitting ensemble simulations were conducted of an isolated, diurnally forced SDC thunderstorm during Cloud, Aerosol, and Complex Terrain Interactions (CACTI)/Remote Sensing of Electrification, Lightning, and Mesoscale/Microscale Processes with Adaptive Ground Observations (RELAMPAGO). The rich observational data facilitated detailed ensemble verification, where dry biases in the surface energy balance and soil moisture were identified. These biases promoted rapid removal of convective inhibition and an early onset of precipitating cells over the SDC that were shallower and weaker than the observed cell. Correction, and then overcorrection, of the soil moisture bias in two successive ensembles was required to rectify the surface energy balance and improve the representation of the SDC cell. Nevertheless, substantial ensemble variability in convective precipitation was found, with some members producing more widespread convection than observed and others producing no deep convection at all. This variability was largely explained by a combination of thermodynamic and dynamic mechanisms, dominated by a positive sensitivity of convective precipitation to preconvective moist instability over the ridge. Secondary sensitivities were found to low-level upward mass flux and midlevel cross-barrier winds, the latter of which caused gravity waves with elevated downdrafts that tended to suppress incipient clouds.

Lopez, Andres [Department of Atmospheric and Ocean↗
DOE OSTI2025

Dataset for "Machine Learning Ensembles Can Enhance Hydrologic Predictions and Uncertainty Quantification" Willard et al. (2025).

This data release provides all data and code used in the paper " "Machine Learning Ensembles Can Enhance Hydrologic Predictions and Uncertainty Quantifications" Willard et al. (2025)" to model stream temperature, evaluate, and assess results. The associated manuscript explores the effect of different ensemble construction techniques across different common machine learning (ML) architectures for predictions in unmonitored basins. Modeling was done using long short-term memory (LSTM), gated recurrent unit (GRU), temporal convolution network (TCN), and extreme gradient boosting (XGBoost) models, and stream site coverage spans 1362 locations across the conterminous United States. The ensemble construction techniques investigated include ensemble by random weight initialization, differing hyperparameters, different random subsets of training data, different subselections of input features, different architectures, and Monte Carlo Dropout. The data is organized into these items items:Code repository and data for the paper " "Machine Learning Ensembles Can Enhance Hydrologic Predictions and Uncertainty Quantifications" Willard et al. (2025).Code: stream_temp_ml_regionalization.zip contains the code repositoryData to run the code:- data_dir.zip -- contains all files that should be moved to the "DATA_DIR" variable defined in the "set_env_vars.sh" script in the code repository- metadata_dir.zip -- contains all files that should be moved to the "METADATA_DIR" variable defined in the "set_env_vars.sh" script in the code repositoryData produced by the code and used in the paper:- outputs_dir.zip - contains model output and results (outputs_dir/results), model weights (outputs_dir/models), and all other outputs used for the paper including feature importances.To cite this code, please use the following BibTeX or MLA entries:bibtex:@misc{willard2025streamensembles,author = {Jared Willard and Charuleka Varadharajan},title = {Dataset for "Machine Learning Ensembles Can Enhance Hydrologic Predictions and Uncertainty Quantification"},year = {2024},doi = {10.15485/2527393},publisher = {ESS-DIVE Repository},url = {https://data.ess-dive.lbl.gov/datasets/doi:10.15485/2527393}}MLA: Willard, Jared, et al. Dataset for "Machine Learning Ensembles Can Enhance Hydrologic Predictions and Uncertainty Quantification". 2025. ESS-DIVE Repository, doi:10.15485/2448016.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Active operator learning with predictive uncertainty quantification for partial differential equations

With the increased prevalence of neural operators being used to provide rapid solutions to partial differential equations (PDEs), understanding the accuracy of model predictions and the associated error levels is necessary for deploying reliable surrogate models in scientific applications. Existing uncertainty quantification (UQ) frameworks employ ensembles or Bayesian methods, which can incur substantial computational costs during both training and inference. Here, we propose a lightweight predictive UQ method tailored for Deep operator networks (DeepONets) that also generalizes to other operator networks. Numerical experiments on linear and nonlinear PDEs demonstrate that the framework’s uncertainty estimates are unbiased and provide accurate out-of-distribution uncertainty predictions with a sufficiently large training dataset. Our framework provides fast inference and uncertainty estimates that can efficiently drive outer-loop analyses that would be prohibitively expensive with conventional solvers. We demonstrate how predictive uncertainties can be used in the context of Bayesian optimization and active learning problems to yield improvements in accuracy and data-efficiency for outer-loop optimization procedures. In the active learning setup, we extend the framework to Fourier Neural Operators (FNO) and describe a generalized method for other operator networks. To enable real-time deployment, we introduce an inference strategy based on precomputed trunk outputs and a sparse placement matrix, reducing evaluation time by more than a factor of five. Our method provides a practical route to uncertainty-aware operator learning in time-sensitive settings.

97 MATHEMATICS AND COMPUTING↗