Search NASA⌕ Search

SEARCH · Search NASA

Results for “computer system benchmarking”

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 1,567 records · Page 87

Realization of fermionic Laughlin state on a quantum processor

Strongly correlated topological phases of matter are central to modern condensed matter physics and quantum information technology but often challenging to probe and control in material systems. The experimental difficulty of accessing these phases has motivated the use of engineered quantum platforms for simulation and manipulation of exotic topological states. Among these, the Laughlin state stands as a cornerstone for topological matter, embodying fractionalization, anyonic excitations, and incompressibility. Although its bosonic analogs have been realized on programmable quantum simulators, a genuine fermionic Laughlin state has yet to be demonstrated on a quantum processor. Here, we realize the ν = 1/3 fermionic Laughlin state on IonQ’s trapped-ion quantum computer using an efficient and scalable Hamiltonian variational ansatz with 369 two-qubit gates on a 16-qubit circuit. Employing symmetry-verification error mitigation, we extract key observables that characterize the Laughlin state, including correlation hole, bulk-edge correspondence, and topological entanglement entropy, with strong agreement to exact diagonalization benchmarks. This work demonstrates an end-to-end workflow to simulate material-intrinsic topological orders and provides a starting point to explore its dynamics and excitations on digital quantum processors.

Shen, Lingnan [Univ. of Washington, Seattle, WA (U↗

Exploiting a derivative discontinuity estimate for accurate G0W0 ionization potentials and electron affinities

Abstract The GW approximation has become an important tool for predicting charged excitations of isolated molecules and condensed systems. Its popularity can be attributed to many factors, including a favorable scaling and relatively good accuracy. In practical applications, the GW is often performed as a one-shot perturbation known as G 0 W 0 . Unfortunately, G 0 W 0 suffers from a strong starting point dependence and is often not as accurate as one would need. Self-consistent GW methodologies alleviate these problems but come with a marked increase in computational cost. In this manuscript, we propose the use of an estimate of the exchange-correlation derivative discontinuity to provide a remarkably good starting point for G 0 W 0 calculations, yielding ionization potentials and electron affinities with eigenvalue self-consistent GW quality at no additional cost. We assess the quality of the resulting methodology with the GW 100 benchmark set and compare its advantages over other similar methods.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

SharP Data Constructs: Data Constructs to Enable Data-Centric Computing

Extreme-scale applications (i.e., Big-Compute) are becoming increasingly data-intensive, i.e., producing and consuming increasingly large amounts of data. The HPC systems traditionally used for these applications are now used for Big-Data applications such as data analytics, social network analysis, machine learning, and genomics. As a consequence of these trends, the system architecture should be flexible and data-centric. This can already be witnessed in the pre-exascale systems with TBs of on-node hierarchical and heterogeneous memories, PBs of system memory, low-latency, high-throughput networks, and many threaded cores. As such, the pre-exascale systems suit the needs of both Big-Compute and Big-Data applications. Though the system architecture is flexible enough to support both Big-Compute and Big-Data, we argue there is a software gap. Particularly, we need data-centric abstractions to leverage the full potential of the system, i.e., there is a need for native support for data resilience, the ability to express data locality and affinity, mechanisms to reduce data movement, the ability to share data, and abstractions to express User's data usage and data access patterns. In this paper, we (i) show the need for taking a holistic approach towards data-centric abstractions, (ii) show how these approaches were realized in the SHARed data-structure centric Programming abstraction (SharP) library, a data-structure centric programming abstraction, and (iii) apply these approaches to a variety of applications that demonstrate its usefulness. Particularly, we apply these approaches to QMCPack and the Graph500 benchmark and demonstrate the advantages of this approach on extreme-scale systems.

Aderholdt, William Ferrol↗

ENDF/B-VIII.0 Testing With AMPX and SCALE

As the new ENDF/B-VIII.0 data has continued development from its initial beta release into the final release version, it has subsequently undergone processing and testing at Oak Ridge National Laboratory with AMPX and SCALE. Each of the beta releases has been processed with AMPX to create both multigroup (MG) and continuous-energy (CE) libraries. The MG libraries were all processed using the SCALE 252-group structure. These libraries were then used with the VALID suite for benchmarking the new data library’s performance against the currently distributed ENDF/B-VII.1 (VII.1)-based SCALE libraries. Here, we compare the MG and CE VALID results for the two current ENDF releases, versions VII.1 and the most recent beta release of ENDF/B-VIII.0 as of this writing (VIII.0.β4). Generally, both the MG and CE VALID aggregate results for VIII.0.β4 compare reasonably well with their VII.1-derived counterparts. As can be expected of any new ENDF release, some of the individual cases have substantially different computed neutron multiplicity (k eff ) values between the two releases, although typically the differences are less than 100 percent-milli (pcm). However, changes to the plutonium cross sections have induced much larger changes in the ratio of the computed k eff to the experimental value (C/E) for thermal spectrum plutonium solution critical systems. This paper details the comparative performance of the VII.1 and VIII.0.β4 data for the broad range of test cases covered by the VALID suite. A companion paper at this conference provides details of the baseline validation of SCALE 6.2, including the first release of the VII.1 data.

97 MATHEMATICS AND COMPUTING↗

Analysis of Shutdown Dose Rate in Fusion Energy Systems Using Hybrid Monte Carlo/Deterministic Techniques - Paper 138

Accurate assessments of shutdown dose rate (SDDR) are critical to support operation, maintenance, and waste disposal planning and to guide possible design changes of critical components in fusion energy systems. An SDDR calculation involves three steps: 1. a neutron transport calculation to determine the space and energy neutron flux distributions, 2. an activation calculation for computing the photon source distribution, and 3. a photon transport calculation for the estimation of the final SDDR. A companion paper describes the challenges of SDDR computations in fusion energy systems and the status of the techniques that are currently being used in SDDR analysis. Of these techniques, the rigorous 2-step (R2S) computational system entails Monte Carlo (MC) neutron and photon transport calculations coupled with a comprehensive activation step using a dedicated inventory code and library. The use of global MC variance reduction techniques was suggested for accelerating the SDDR MC neutron transport calculation. These techniques, which attempt to calculate MC tallies with nearly uniform relative uncertainties in both the low-flux space-energy regions and the high-flux space-energy regions, do not preferentially focus the MC computational efforts toward space-energy regions of high importance to the final decay dose. The ability of these approaches to accurately predict SDDR is inhibited by their prohibitive computational costs, which will be on the order of thousands of processor-years for full-scale modeling of an entire fusion plant. The companion paper describes the theoretical background of the Multi-Step Consistent Adjoint Driven Importance Sampling (MS-CADIS) method, which has been proposed to speed up SDDR MC neutron calculations. The MS-CADIS method uses the CADIS method, which has been successfully used for more than a decade in shielding calculations, but focuses on multistep shielding calculations such as SDDR analyses. The companion paper also describes a new method for calculating the uncertainties in SDDR due to uncertainties in the MC neutron calculation. This new method of uncertainty propagation uses the MS-CADIS neutron adjoint source to propagate the uncertainties in the neutron fluxes to the SDDR. This paper describes the application of the MS-CADIS method for speeding up the SDDR MC calculations and for calculating the SDDR uncertainty due to the neutron flux uncertainties. A new metric for assessing the reliability of SDDR calculations was also suggested and used in this analysis. The ITER benchmark problem was used in this analysis. The problem resembles the configuration and geometrical arrangement of an upper port plug in ITER. The analysis compared the efficiency of the MS-CADIS method to the traditional approach of using global MC variance reduction techniques for speeding up SDDR MC neutron calculation. The application of the MS-CADIS method to SDDR calculations in fusion energy systems was tested using the ITER benchmark problem. Compared to the standard FW-CADIS method, the increase in the efficiency of the SDDR neutron MC calculation due to the use of the MS-CADIS method was between 18% and 69%. The MS-CADIS method also increases the fraction of non-zero-scoring mesh tally elements in the space-energy regions of high importance to the final SDDR. Implementation of the MS-CADIS method in the SCALE and the ADVANTG code systems is currently under way. (authors)

61 RADIATION PROTECTION AND DOSIMETRY↗

Investigation of Stress Field and Fracture Development During Shale Maturation Using Analog Rock Systems

The emergence of hydrocarbons within shale as a major recoverable resource has sparked interest in fluid transport through these tight mudstones. Recent studies suggest the importance to recovery of microfracture networks that connect localized zones with large organic content to the inorganic matrix. This paper presents a joint modeling and experimental study to examine the onset, formation, and evolution of microfracture networks as shale matures. Both the stress field and fractures are simulated and imaged. A novel laboratory-scale, phase-field fracture propagation model was developed to characterize the material failure mechanisms that play a significant role during the shale maturation process. The numerical model developed consists of coupled solid deformation, pore pressure, and fracture propagation mechanisms. Benchmark tests were conducted to validate model accuracy. Laboratory-grade gelatins with varying Young’s modulus were used as scaled-rock analogs in two dimensional Hele-Shaw cell apparatus. Yeast within the gelatin generates gas in a fashion analogous to hydrocarbon formation as shale matures. These setups allow study and visualization of host rock elastic-brittle fracture and fracture network propagation mechanisms. The experimental setup was fitted to utilize photoelasticity principles coupled with birefringence properties of gelatin to explore visually the stress field of the gelatin as the fracture network developed. Stress optics image analysis and Linear Elastic Fracture Mechanics (LEFM) principles for crack propagation were used to monitor fracture growth for each gelatin type. Observed and simulated responses suggest gas diffusion within and deformation of the gelatin matrix as predominant mechanisms for energy dissipation depending on gelatin strength. LEFM, an experimental estimation of principal stress development with fracture growth, at different stages was determined for each gelatin rheology. Here, the interplay of gas diffusion and material deformation determines the resulting frequency and pattern of fractures. Results correlate with Young’s modulus. Experimental and computed stress fields reveal that fractures resulting from internal gas generation are similar to, but not identical to, type 1 opening mode.

58 GEOSCIENCES↗

Ansatz-Free Hamiltonian Learning with Heisenberg-Limited Scaling

Learning the unknown interactions that govern a quantum system is crucial for quantum information processing, device benchmarking, and quantum sensing. The problem, known as Hamiltonian learning, is well understood under the assumption that interactions are local, but this assumption may not hold for arbitrary Hamiltonians. Previous methods all require high-order inverse polynomial dependency with precision, unable to surpass the standard quantum limit and reach the gold-standard Heisenberg-limited scaling. Whether Heisenberg-limited Hamiltonian learning is possible without prior assumptions about the interaction structures, a challenge we term ansatz-free Hamiltonian learning , remains an open question. In this work, we present a quantum algorithm to learn arbitrary sparse Hamiltonians without any structure constraints using only black-box queries of the system’s real-time evolution and minimal digital controls to attain Heisenberg-limited scaling in estimation error. Our method is also resilient to state-preparation-and-measurement errors, enhancing its practical feasibility. We numerically demonstrate our ansatz-free protocol for learning physical Hamiltonians and validating analog quantum simulations, benchmarking our performance against the state-of-the-art Heisenberg-limited learning approach. Moreover, we establish a fundamental trade-off between total evolution time and quantum control on learning arbitrary interactions, revealing the intrinsic interplay between controllability and total evolution-time complexity for any learning algorithm. These results pave the way for further exploration into Heisenberg-limited Hamiltonian learning in complex quantum systems under minimal assumptions, potentially enabling new benchmarking and verification protocols.

machine learning↗

Thermal-striping analysis methodology for sodium-cooled reactor design

Computational fluid dynamics simulations have been performed to study the applicability of engineering CFD methods for thermals striping analysis. Thermal striping is the fluctuating temperature profile in a solid caused by fluctuating fluid temperature, and the resulting fluctuating thermal stresses cause high cycle fatigue and eventual material failure. Here this study presents the methodology for thermal striping analysis that including a transient conjugate heat transfer RANS model and benchmark the results against a sodium triple jet experiment. The multiphysics calculations are implemented to analyze the thermal stresses in the solid domain caused by the coupled heat transfer between the fluid and solid domains. Validation data from a liquid sodium triple jet experiment include time-averaged temperature measurements and power spectra of the temperature signal. The numerical results agree well with these experimental measures, demonstrating key features such as the dominant frequency of temperature fluctuations. The applicability of a low-cost wall treatment method is demonstrated, enabling key computational savings. Finally, the performance of two finite element stress analysis software packages is compared, and the validity of the lower-cost method is confirmed. These results demonstrate the applicability of engineering methods for computational thermal striping calculations, enabling thermal striping estimations in large fluid systems such as the core of a Sodium-cooled nuclear reactor.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Quantifying uncertainty for deep learning based forecasting and flow-reconstruction using neural architecture search ensembles

Classical problems in computational physics such as data-driven forecasting and signal reconstruction from sparse sensors have recently seen an explosion in deep neural network (DNN) based algorithmic approaches. However, most DNN models do not provide uncertainty estimates, which are crucial for establishing the trustworthiness of these techniques in downstream decision making tasks and scenarios. In recent years, ensemble-based methods have achieved significant success for the uncertainty quantification in DNNs on a number of benchmark problems. However, their performance on real-world applications remains under-explored. In this work, we present an automated approach to DNN discovery and demonstrate how this may also be utilized for ensemble-based uncertainty quantification. Specifically, we propose the use of a scalable neural and hyperparameter architecture search for discovering an ensemble of DNN models for complex dynamical systems. We highlight how the proposed method not only discovers high-performing neural network ensembles for our tasks, but also quantifies uncertainty seamlessly. This is achieved by using genetic algorithms and Bayesian optimization for sampling the search space of neural network architectures and hyperparameters. Subsequently, a model selection approach is used to identify candidate models for an ensemble set construction. Afterwards, a variance decomposition approach is used to estimate the uncertainty of the predictions from the ensemble. We demonstrate the feasibility of this framework for two tasks — forecasting from historical data and flow reconstruction from sparse sensors for the sea-surface temperature. In conclusion, we demonstrate superior performance from the ensemble in contrast with individual high-performing models and other benchmarks.

Deep ensembles↗

Deep Learning enabled spectral energy conversion for in situ exposure measurements

A detector-specific deep learning (DL) approach is presented for spectra-to-exposure conversion using large-format sodium iodide (NaI(Tl)) detectors deployed for in situ environmental radiation measurements in emergency response scenarios. Accurate determination of exposure from NaI spectra is challenging due to poor energy resolution, partial energy absorption, and the strong sensitivity of traditionally deployed analytical conversion methods to calibrated source geometry and pre-deployment assumptions. Here, to address these limitations, a multi-layer perceptron model was trained on a hybrid in situ /Monte Carlo dataset constructed to span a broad range of photon energies, spatial extents, and realistic deployment variability, representative of general in situ emergency response conditions. The DL model was evaluated against commonly fielded analytical approaches under matched simulation conditions, including a single-factor method, a G-function method, and a modeled pressurized ion chamber (PIC) baseline. This study was intentionally computational in scope to enable controlled, like-for-like comparisons between conversion techniques while minimizing confounding real-world variability. Comparison to the modeled PIC provides contextual benchmarking and is not intended as a field inter-comparison with deployed instruments. Across the evaluated 20 keV to 3 MeV energy range, the DL approach consistently exhibited higher accuracy and reduced variance relative to the analytical methods against a deterministically calculated exposure. This may indicate improved robustness to spectral complexity without reliance on source-, geometric-, or spectral region-specific optimization. While results do not represent real-world validation, the presented work demonstrates that deep learning may effectively learn the nonlinear detector response-to-exposure relationship for asymmetric NaI(Tl) detectors and offers a promising pathway for improving in situ exposure estimation using spectroscopic systems already integrated into initial real-time emergency response operations.

61 RADIATION PROTECTION AND DOSIMETRY↗

ExaAM: Metal additive manufacturing simulation at the fidelity of the microstructure

Additive manufacturing (AM), or 3D printing, of metals is transforming the fabrication of components, in part by dramatically expanding the design space, allowing optimization of shape and topology. However, although the physical processes involved in AM are similar to those of welding, a field with decades of experimental, modeling, simulation, and characterization experience, qualification of AM parts remains a challenge. The availability of exascale computational systems, particularly when combined with data-driven approaches such as machine learning, enables topology and shape optimization as well as accelerated qualification by providing process-aware, locally accurate microstructure and mechanical property models. We describe the physics components comprising the Exascale Additive Manufacturing simulation environment and report progress using highly resolved melt pool simulations to inform part-scale finite element thermomechanics simulations, drive microstructure evolution, and determine constitutive mechanical property relationships based on those microstructures using polycrystal plasticity. We report on implementation of these components for exascale computing architectures, as well as the multi-stage simulation workflow that provides a unique high-fidelity model of process–structure–property relationships for AM parts. In addition, we discuss verification and validation through collaboration with efforts such as AM-Bench, a set of benchmark test problems under development by a team led by the National Institute of Standards and Technology.

3D printing↗

miniGMG

miniGMG is a compact geometric multigrid (MG benchmark designed to proxy the performance characteristics of the solves found in adaptive mesh refinement multigrid (AMR MG) applications. It solves the equation a*alpha*u - b div beta grad u = f on a large cubical domain using cell-centered values and periodic boundaries. a and b are scalar constants, alpha and beta are space varying constants, f is the right hand side, and u is the solution. The cubical domain is divided into cubical subdomains which are distributed across the supercomputer. The righthand side f is generated from a continuous function for u. Thus, one can solve this linear system of equations and verify the error properties of the discrete solution u by comparing to a sampling of the continuous u. miniGMG implements both v- and f- cycles, as well as two constant coeffficient discretications of the equation above (one 27pt 2nd order, and one 13pt 4th order). In both cases, the (embedded) v-cycles are truncated when a subdomain reaces 4^3, 2^3, or 1^3 cells (a u-cycle). The multigrd solver then swtiches to one of seven coarse grid (bottom) solves. relaxation, BiCGStab, CABiCGStab, TelescopingCABiCGStab, CG, or CACG. The code exploits two forms of communication-avoiding. The first is exemplified by the referenced SC'12 paper in which DRAM data movment is avoided when smoothing. The second, described in the IPDPS'14 paper avoids collectives in a bottom solve (the CA bottom solves). There are a number of implementations of the operators for both CPUs and NVIDIA GPUs. The CPUs have been optimized to exploit a wavefront approach to communication-avoiding. The GPU versions include a number of optimized and tunable implementations. The code has been demonstrated to scale to 46K nodes on the BGQ Mira (750K cores), 9K nodes on Edison (111K cores), and 14K GPUs on Titan (37M cuda cores). It thus acts as a scalable testbed for a wide range of computer science and applied math research. Nominally, one invokes the benchmark as ./run b x y z alpha beta gamma Each subdomain is (2^b)^3 cells Each MPI process receives a x-by-y-by-z collection of subdomains. There are alpha*beta*gamma total MPI processes For correctness (a cubical domain), *alpha==y*beta==z*gamma.

Williams, Sameul↗

GradDFT. A software library for machine learning enhanced density functional theory

Density functional theory (DFT) stands as a cornerstone method in computational quantum chemistry and materials science due to its remarkable versatility and scalability. Yet, it suffers from limitations in accuracy, particularly when dealing with strongly correlated systems. To address these shortcomings, recent work has begun to explore how machine learning can expand the capabilities of DFT: an endeavor with many open questions and technical challenges. In this work, we present GradDFT a fully differentiable JAX-based DFT library, enabling quick prototyping and experimentation with machine learning-enhanced exchange–correlation energy functionals. GradDFT employs a pioneering parametrization of exchange–correlation functionals constructed using a weighted sum of energy densities, where the weights are determined using neural networks. Moreover, GradDFT encompasses a comprehensive suite of auxiliary functions, notably featuring a just-in-time compilable and fully differentiable self-consistent iterative procedure. To support training and benchmarking efforts, we additionally compile a curated dataset of experimental dissociation energies of dimers, half of which contain transition metal atoms characterized by strong electronic correlations. The software library is tested against experimental results to study the generalization capabilities of a neural functional across potential energy surfaces and atomic species, as well as the effect of training data noise on the resulting model accuracy.

Chemistry↗

SU-F-T-231: Improving the Efficiency of a Radiotherapy Peer-Review System for Quality Assurance

Purpose: To improve the efficiency of a radiotherapy peer-review system using a commercially available software application for plan quality evaluation and documentation. Methods: A commercial application, FullAccess (Radialogica LLC, Version 1.4.4), was implemented in a Citrix platform for peer-review process and patient documentation. This application can display images, isodose lines, and dose-volume histograms and create plan reports for peer-review process. Dose metrics in the report can also be benchmarked for plan quality evaluation. Site-specific templates were generated based on departmental treatment planning policies and procedures for each disease site, which generally follow RTOG protocols as well as published prospective clinical trial data, including both conventional fractionation and hypo-fractionation schema. Once a plan is ready for review, the planner exports the plan to FullAccess, applies the site-specific template, and presents the report for plan review. The plan is still reviewed in the treatment planning system, as that is the legal record. Upon physician’s approval of a plan, the plan is packaged for peer review with the plan report and dose metrics are saved to the database. Results: The reports show dose metrics of PTVs and critical organs for the plans and also indicate whether or not the metrics are within tolerance. Graphical results with green, yellow, and red lights are displayed of whether planning objectives have been met. In addition, benchmarking statistics are collected to see where the current plan falls compared to all historical plans on each metric. All physicians in peer review can easily verify constraints by these reports. Conclusion: We have demonstrated the improvement in a radiotherapy peer-review system, which allows physicians to easily verify planning constraints for different disease sites and fractionation schema, allows for standardization in the clinic to ensure that departmental policies are maintained, and builds a comprehensive database for potential clinical outcome evaluation.

60 APPLIED LIFE SCIENCES↗

Patch Hierarchical Attention Transformer for Efficient Particle Jet Tagging

Real-time jet tagging is critical for identifying short-lived particle decays in the high-throughput detectors of the Large Hadron Collider, where real-time trigger systems responsible for deciding which collision events to store impose strict latency and accuracy constraints. While transformer architectures achieve the highest jet tagging accuracy when compute is unconstrained, their quadratic self-attention cost makes inference restrictive on trigger budget. Existing efficient variants reduce the computational cost, but hinder the classification performance. To address this limitation, we introduce the Patch Hierarchical Attention Transformer (PHAT-JeT), which combines two mechanisms: a physics-inspired geometric message-passing module that encodes local detector-plane structure, and a hierarchical patch-based attention scheme that computes exact attention within small particle groups while preserving global context through lightweight patch-token communication. Within a restricted budget, PHAT-JeT achieves state-of-the-art accuracy and background rejection among all resource-constrained jet tagging models on four benchmarks (\textsc{hls4ml}, JetClass, Top Tagging, and Quark--Gluon). Our code is available at https://github.com/aaronw5/PHAT-JeT.

Wang, Aaron [Illinois U., Chicago] (ORCID:00000003↗

Calculation of natural convection test at Phenix using the NETFLOW++ code

The present paper describes modeling and analyses of a natural convection of the pool-type fast breeder reactor Phenix. The natural convection test was carried out as one of the End of Life Tests of the Phenix. Objective of the present study is to assess the applicability of the NETFLOW++ code which has been verified thus far using various water facilities and validated using the plant data of the loop-type FBR 'Monju' and the loop-type experimental fast reactor 'Joyo'. The Phenix primary heat transport system is modeled based on the benchmark documents available from IAEA. The calculational model consists of only the primary heat transport system with boundary conditions on the secondary-side of IHX. The coolant temperature at the primary pump inlet, the primary coolant temperature at the IHX inlet and outlet, the secondary coolant temperatures and other parameters are calculated by the code where the heat transfer between the hot and cold pools is explicitly taken into account. A model including the secondary and tertiary systems was prepared, and the calculated results also agree well with the measured data in general. (authors)

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

Bayesian Optimized Deep Ensemble for Uncertainty Quantification of Deep Neural Networks: a System Safety Case Study on Sodium Fast Reactor Thermal Stratification Modeling

Deep neural networks (DNNs) are increasingly important to scientific computing and engineering system simulations. Accurate uncertainty quantification (UQ) for DNNs is critical in safety-sensitive engineering domains. Traditional Deep Ensemble (DE) methods, while easy to implement, frequently suffer from poorly calibrated uncertainty estimates and limited predictive accuracy due to reliance on fixed architectures with varied weight initializations. To address these issues, we introduce a workflow that combines Bayesian Optimization (BO) and DE. The workflow is modular, scalable, and integrates parallel BO initialized with Sobol sequences to individually optimize the hyperparameters of each ensemble member. This method enhances ensemble diversity, improves predictive accuracy, and provides reliable uncertainty estimates. We evaluate the proposed BODE approach in a sodium fast reactor thermal stratification modeling case study, where we used a densely connected convolutional neural network to predict turbulent viscosity during the reactor transient with consideration of data noise. We benchmark its performance against several optimization approaches, including baseline deep ensemble, evolutionary algorithm-optimized ensemble, ensemble formed via random search combined with greedy selection, and a BO ensemble using random initialization. Here, our results demonstrate superior performance of the developed BODE approach. In noise-free scenarios, BODE notably reduces incorrect aleatoric uncertainty and significantly enhances predictive accuracy. Under conditions of 5% and 10% Gaussian noise, BODE adaptively quantifies uncertainty proportional to data noise, achieving up to an 80% reduction in root mean square error compared to baseline methods and producing well-calibrated prediction intervals.

Bayesian optimization↗