An official website of the United States government
Abstract Subgrid‐scale processes, such as atmospheric gravity waves (GWs), play a pivotal role in shaping the Earth's climate but cannot be explicitly resolved in climate models due to limitations on resolution. Instead, subgrid‐scale parameterizations are used to capture their effects. Recently, machine learning (ML) has emerged as a promising approach to learn parameterizations. In this study, we explore uncertainties associated with a ML parameterization for atmospheric GWs. Focusing on the uncertainties in the training process (parametric uncertainty), we use an ensemble of neural networks to emulate an existing GW parameterization. We estimate both offline uncertainties in raw NN output and online uncertainties in climate model output, after the neural networks are coupled. We find that online parametric uncertainty contributes a significant source of uncertainty in climate model output that must be considered when introducing NN parameterizations. This uncertainty quantification provides valuable insights into the reliability and robustness of ML‐based GW parameterizations, thus advancing our understanding of their potential applications in climate modeling.
more »
« less
Quantifying Uncertainty in Ensemble Deep Learning
Neural networks are an emerging topic in the data science industry due to their high versatility and efficiency with large data sets. Past research has utilized machine learning on experimental data in the material sciences and chemistry field to predict properties of metal oxides. Neural networks can determine underlying optical properties in complex images of metal oxides and capture essential features which are unrecognizable by observation. However, neural networks are often referred to as a “black box algorithm” due to the underlying process during the training of the model. This poses a concern on how robust and reliable the prediction model actually is. To solve this ensemble neural networks were created. By utilizing multiple networks instead of one the robustness of the model was increased and points of uncertainty were identified. Overall, ensemble neural networks outperform singular networks and demonstrate areas of uncertainty and robustness in the model.
more »
« less
- Award ID(s):
- 2050754
- PAR ID:
- 10520058
- Publisher / Repository:
- SIAM
- Date Published:
- Journal Name:
- SIAM undergraduate research online
- Volume:
- 16
- ISSN:
- 2327-7807
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Deep learning has emerged as a promising paradigm to give access to highly accurate predictions of molecular and material properties. A common short-coming shared by current approaches, however, is that neural networks only give point estimates of their predictions and do not come with predictive uncertainties associated with these estimates. Existing uncertainty quantification efforts have primarily leveraged the standard deviation of predictions across an ensemble of independently trained neural networks. This incurs a large computational overhead in both training and prediction, resulting in order-of-magnitude more expensive predictions. Here, we propose a method to estimate the predictive uncertainty based on a single neural network without the need for an ensemble. This allows us to obtain uncertainty estimates with virtually no additional computational overhead over standard training and inference. We demonstrate that the quality of the uncertainty estimates matches those obtained from deep ensembles. We further examine the uncertainty estimates of our methods and deep ensembles across the configuration space of our test system and compare the uncertainties to the potential energy surface. Finally, we study the efficacy of the method in an active learning setting and find the results to match an ensemble-based strategy at order-of-magnitude reduced computational cost.more » « less
-
Chemical reaction neural network (CRNN), a recently developed tool for autonomous discovery of reaction models, has been successfully demonstrated on a variety of chemical engineering and biochemical systems. It leverages the extraordinary data-fitting capacity of modern deep neural networks (DNNs) while preserving high interpretability and robustness by embedding widely applicable physical laws such as the law of mass action and the Arrhenius law. In this paper, we further developed Bayesian CRNN to not only reconstruct but also quantify the uncertainty of chemical kinetic models from data. Two methods, the Markov chain Monte Carlo algorithm and variational inference, were used to perform the Bayesian CRNN, with the latter mainly adopted for its speed. We demonstrated the capability of Bayesian CRNN in the kinetic uncertainty quantification of different types of chemical systems and discussed the importance of embedding physical laws in data-driven modeling. Finally, we discussed the adaptation of Bayesian CRNN for incomplete measurements and model mixing for global uncertainty quantification.more » « less
-
null (Ed.)In this study, we show how strong metal–support interaction (SMSI) oxides in Pt–Nb/SiO 2 and Pt–Ti/SiO 2 affect the electronic, geometric and catalytic properties for propane dehydrogenation. Transmission electron microscopy (TEM), CO chemisorption, and decrease in the catalytic rates per gram Pt confirm that the Pt nanoparticles were partially covered by the SMSI oxides. X-ray absorption near edge structure (XANES), in situ X-ray photoelectron spectroscopy (XPS), and resonant inelastic X-ray scattering (RIXS) showed little change in the energy of Pt valence orbitals upon interaction with SMSI oxides. The catalytic activity per mol of Pt for ethylene hydrogenation and propane dehydrogenation was lower due to fewer exposed Pt sites, while turnover rates were similar. The SMSI oxides, however, significantly increase the propylene selectivity for the latter reaction compared to Pt/SiO 2 . In the SMSI catalysts, the higher olefin selectivity is suggested to be due to the smaller exposed Pt ensemble sites, which result in suppression of the alkane hydrogenolysis reaction; while the exposed atoms remain active for dehydrogenation.more » « less
-
Abstract Chromophoric dissolved organic matter (CDOM) is an important part of ocean carbon biogeochemistry with relevance to long‐term observations of ocean biology due to its dominant light absorption properties. Thus, understanding the underlying processes controlling CDOM distribution is important for predicting changes in light availability, primary production, and the cycling of biogeochemically important matter. We present a biogeochemical CDOM model for the open ocean with two classes of biological lability and uncertainty estimates derived from 43 ensemble members that provide a range of model parameter variations. Ensemble members were optimized to match global ocean in situ CDOM measurements and independently assessed against satellite CDOM estimates, which showed good agreement in spatial patterns. Based on the ensemble median, we estimate that about 7% of open‐ocean CDOM is of terrestrial origin, but the ensemble range is large (<0.1–26%). CDOM is rapidly removed in the surface ocean (<200 m) due to biological degradation for short‐lived CDOM and photodegradation for long‐lived CDOM, leading to a net flux of CDOM to the surface ocean from the dark ocean. This deep‐water source (ensemble median 0.001 m−1 yr−1) is similar in magnitude to the riverine flux (0.005 m−1 yr−1) into the surface ocean. Though discrepancies between the model and observational data remain, this work serves as a foundational framework for a mechanistic assessment of global CDOM distribution that is independent of satellite data.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- The DOI is not currently available.
