Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “bayesian optimization”

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 667 records · Page 37

Marginal unbiased score expansion and application to CMB lensing

Here, we present the marginal unbiased score expansion (MUSE) method, an algorithm for generic high-dimensional hierarchical Bayesian inference. MUSE performs approximate marginalization over arbitrary non-Gaussian latent parameter spaces, yielding Gaussianized asymptotically unbiased and near-optimal constraints on global parameters of interest. It is computationally much cheaper than exact alternatives like Hamiltonian Monte Carlo (HMC), excelling on funnel problems which challenge HMC, and does not require any problem-specific user supervision like other approximate methods such as variational inference or many simulation-based inference methods. MUSE makes possible the first joint Bayesian estimation of the delensed Cosmic Microwave Background (CMB) power spectrum and gravitational lensing potential power spectrum, demonstrated here on a simulated data set as large as the upcoming South Pole Telescope 3G 1500 deg 2 survey, corresponding to a latent dimensionality of ~6 million and of order 100 global bandpower parameters. On a subset of the problem where an exact but more expensive HMC solution is feasible, we verify that MUSE yields nearly optimal results. We also demonstrate that existing spectrum-based forecasting tools which ignore pixel-masking underestimate predicted error bars by only ~10%. This method is a promising path forward for fast lensing and delensing analyses which will be necessary for future CMB experiments such as SPT-3G, Simons Observatory, or CMB-S4, and can complement or supersede existing HMC approaches. The success of MUSE on this challenging problem strengthens its case as a generic procedure for a broad class of high-dimensional inference problems.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Bayesian inferences of electrical current delivered to shocked transmission lines

Small radius Bdot measurements of electrical current delivered by pulsed power devices are routinely compromised by electrode/convolute plasmas endemic to multiterawatt transmission lines. Inferences of delivered current unaffected by these issues have recently been obtained by numerically optimizing consistency between model predictions and local experimental velocimetry data, but these are only unique for shockless velocity profiles. In the paper we describe a more general Bayesian method capable of inferring current despite the presence of shocks. Additionally, we describe uncertainty estimates and use of the technique on experimental data. This technique is the first to provide uncertainty estimates on the full current trace delivered to an inertial confinement fusion target.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Learning Weight Uncertainty with Stochastic Gradient MCMC for Shape Classification

Learning the representation of shape cues in 2D & 3D objects for recognition is a fundamental task in computer vision. Deep neural networks (DNNs) have shown promising performance on this task. Due to the large variability of shapes, accurate recognition relies on good estimates of model uncertainty, ignored in traditional training of DNNs, typically learned via stochastic optimization. This paper leverages recent advances in stochastic gradient Markov Chain Monte Carlo (SG-MCMC) to learn weight uncertainty in DNNs. It yields principled Bayesian interpretations for the commonly used Dropout/DropConnect techniques and incorporates them into the SG-MCMC framework. Extensive experiments on 2D & 3D shape datasets and various DNN models demonstrate the superiority of the proposed approach over stochastic optimization. Our approach yields higher recognition accuracy when used in conjunction with Dropout and Batch-Normalization.

convolutional neural network↗

Preface to the Special Issue on TOUGH Symposium 2015

The TOUGH Symposium 2015 was held in Berkeley, California, September 28-30, 2015. The TOUGH family of codes, developed at the Energy Geosciences Division of Lawrence Berkeley National Laboratory (LBNL), is a suite of computer programs for the simulation of multiphase and multicomponent fluid and heat flows in porous and fractured media with applications in many geosciences fields, such as geothermal reservoir engineering, nuclear waste disposal, geological carbon sequestration, oil and gas reservoirs, gas hydrate research, vadose zone hydrology and environmental remediation. Since the first release in the 1980s, many modifications and enhancements have been continuously made to TOUGH and its various descendants (iTOUGH2, TOUGH+, TOUGH-MP, TOUGHREACT, TOUGH+HYDRATE, TMVOC...), at LBNL and elsewhere. Today, these codes are used worldwide in academia, government organizations and private companies in problems involving coupled hydrological, thermal, biogeochemical and geomechanical processes. The Symposia, organized every 2-3 years, bring together developers and users for an open exchange on recent code enhancements and applications. In 2015, the Symposium was attended by one hundred participants, representing thirty-four nationalities. This Special Issue in Computers & Geosciences gathers extended versions of selected Symposium proceedings related to (i) recent enhancements to the TOUGH family of codes and (ii) coupled flow and geomechanics processes modeling. In this special issue, a paper by Jung et al. provides a new base version of TOUGH, TOUGH3. This latest version addresses the computational robustness and efficiency needed for the increasing complexity of the simulated processes and the growing size of model domains. Moreover, it consolidates the serial and parallel versions into a single code, hence, inconsistencies between different versions are less likely to occur, and code maintenance and development become more efficient. Finsterle et al. present a review of the capabilities of iTOUGH2 and illustrate the workflow of an integrated simulation-optimization framework. Liu et al. present an approach to reduce the computational cost of performing Bayesian inverse modeling and demonstrate their approach using a synthetic experiment simulated with TOUGH2. Regarding the mesh generation process, the paper by Doughty presents a generalization of the grid-generation algorithm embedded in TOUGH2 (MESHMAKER) to create one-column grids with fractal dimension, useful for representing flow through fractures or highly heterogeneous media with fractal geometry, and also for inverse methods. Sentís and Gable present a module to interface the open source code Lagrit with TOUGH2, that assists in the processes of mesh generation and model setup, and that can be particularly useful for complex geometries and large models. Bonduà et al. present a pre-processing tool (VORO2MESH) that features a full 3D Voronoi approach for structured and unstructured grids, and an enhanced post-processor (TOUGH2Viewer) for the visualization of 3D Voronoi grids. The literature review shows that coupled fluid flow and geomechanics has gained increasing attention in the last years in many geosciences fields. In this context, several codes of the TOUGH family have been successfully coupled with geomechanics software. The paper by Rutqvist presents a review of some of these TOUGH-based geomechanics models, of which the TOUGH-FLAC simulator – presented for the first time in the TOUGH Symposium 2003 – is well-known worldwide and has been used in a large variety of applications, encompassing the application domains of the TOUGH codes. Blanco-Martín et al. present an extension of TOUGH-FLAC to the finite strain framework, motivated by the existence of large strains associated with rock salt deformation in the context of heat-generating nuclear waste disposal. Following a paper published in the Special Issue of the TOUGH Symposium 2012 in this journal, Kim et al. present the current status of the TOUGH-RSBN simulator for coupled hydro-mechanical modeling of hydraulic fracturing using a discrete fracture network approach, and present some validation and demonstration cases. Rinaldi and Nespoli present TOUGH-SEED, an injection-induced seismicity modeling approach using TOUGH2 and a geomechanics stochastic model, and apply it to the Basel EGS project. Finally, Rinaldi et al. use the PEST protocol implemented in iTOUGH2, combined with TOUGH-FLAC as the forward model, to perform inverse modeling for the In Salah CO2 demonstration site in Algeria. Using pore pressures and InSAR displacements as observations, this approach allows for a multi-physics joint inversion and takes advantage of the inverse capabilities of iTOUGH2 (parameter estimation, sensitivity analysis, uncertainty estimation …), applied to coupled fluid flow and geomechanics. On behalf of the members of the TOUGH Symposium 2015 Organizing Committee, I hope that this collection of papers will provide the readers of Computers & Geosciences with a recent view of the capabilities of the TOUGH family of codes, as well as with current modeling research domains in the geosciences. Last but not least, I would like to express my gratitude to all the reviewers who kindly contributed to this Special Issue, as well as the Journal staff who have made this publication possible.

99 GENERAL AND MISCELLANEOUS↗

Online LIBS–ML Framework for Dynamic Characterization of Heterogeneous Waste-Derived Gasification Feedstocks

LIBS−ML framework for real time feedstock characterization during continuous conveyor transport Heterogeneous waste derived feedstocks (e.g., waste coal, biomass and blends) introduce rapid variability in heating value and ash chemistry that affect gasifier operation, yet conventional laboratory characterization techniques are too slow to support proactive control. To address this gap, this study reports on an online, in situ, dynamic characterization framework that couple’s laser-induced breakdown spectroscopy (LIBS) with leakage safe machine learning (ML) regression to deliver real time, decision quality predictions of gasifier relevant properties. A controlled sample matrix spanning two different waste coals, two different biomasses, and engineered blends under two particle size conditions were constructed and benchmarked using standardized laboratory analyses for proximate/ultimate properties and ash composition. LIBS spectra were acquired dynamically as material flowed on a conveyor belt, using high energy 1064 nm laser ablation and shot averaging to improve repeatability and precision. Supervised regression models (multi layer perceptron (MLP) /artificial neural network (ANN), random forest (RF), and support vector regression (SVR)) and an optimized weighted ensemble were trained on emission line feature sets using nested cross validation with Bayesian hyperparameter tuning and validated against an independent hold out set. The proposed LIBS−ML workflow achieves near laboratory predictive fidelity across parametric targets (including higher heating value (HHV), ash content, fixed carbon, sulfur, major ash forming oxides, and initial deformation temperature (IDT)), with the weighted ensemble providing a robust default predictor under dynamic measurement conditions. These results demonstrate a practical pathway for real time feedstock characterization that can enable feedforward adjustments and more resilient gasifier operation for variable quality waste derived fuels.

Biomass↗

Adaptive hyperparameter updating for training restricted Boltzmann machines on quantum annealers

Restricted Boltzmann Machines (RBMs) have been proposed for developing neural networks for a variety of unsupervised machine learning applications such as image recognition, drug discovery, and materials design. The Boltzmann probability distribution is used as a model to identify network parameters by optimizing the likelihood of predicting an output given hidden states trained on available data. Training such networks often requires sampling over a large probability space that must be approximated during gradient based optimization. Quantum annealing has been proposed as a means to search this space more efficiently which has been experimentally investigated on D-Wave hardware. D-Wave implementation requires selection of an effective inverse temperature or hyperparameter (β) within the Boltzmann distribution which can strongly influence optimization. Here, we show how this parameter can be estimated as a hyperparameter applied to D-Wave hardware during neural network training by maximizing the likelihood or minimizing the Shannon entropy. We find both methods improve training RBMs based upon D-Wave hardware experimental validation on an image recognition problem. Neural network image reconstruction errors are evaluated using Bayesian uncertainty analysis which illustrate more than an order magnitude lower image reconstruction error using the maximum likelihood over manually optimizing the hyperparameter. The maximum likelihood method is also shown to out-perform minimizing the Shannon entropy for image reconstruction.

97 MATHEMATICS AND COMPUTING↗

From Optimization to Sampling Through Gradient Flows

Optimization and sampling algorithms play a central role in science and engineering as they enable finding optimal predictions, policies, and recommendations, as well as expected and equilibrium states of complex systems. The notion of “optimality” is formalized by the choice of an objective function, while the notion of an “expected” state is specified by a probabilistic model for the distribution of states. Optimizing rugged objective functions and sampling multimodal distributions is computationally challenging, especially in high-dimensional problems. Here, for this reason, many optimization and sampling methods have been developed by researchers working in disparate fields such as Bayesian statistics, molecular dynamics, genetics, quantum chemistry, machine learning, weather forecasting, econometrics, and medical imaging.

Trillos, N. García↗

ZEM: Integrated Framework for Real-Time Data and Model Analyses for Robust Environmental Management Decision Making - 16438

Typically environmental management (EM) problems require analysis of large and complex data sets originating from concurrent data streams with different data collection frequencies and pedigree. These big data sets require on-the-fly analysis, assimilation and integration into a series of models with different complexity for various types of model-based analyses. This is needed to provide fast iterative model analyses based on the latest available data to guide decision-making. Furthermore, the data and model are associated with uncertainties. The uncertainties are probabilistic (e.g. measurement errors) and non-probabilistic (unknowns, e.g. alternative conceptual models characterizing site conditions). To address all of these issues, we have developed an integrated framework for real-time data and model analyses for environmental decision-making called ZEM. The framework allows for seamless and on-the-fly integration of data and modeling results for robust and scientifically-defensible decision-making applying advanced decision analyses tools such as Bayesian-Information-Gap Decision Theory (BIG-DT). The framework also includes advanced methods for optimization that are capable of dealing with a large number of unknown model parameters, and surrogate (reduced order) modeling capabilities based on support vector regression techniques. The framework is predominantly coded in Julia, a state-of-the-art high-performance programing language (http://julialang.org) and incorporates the MADS computational tools (Model Analysis and Decision Support; http://mads.lanl.gov). We demonstrate the application of ZEM for data and model analyses at the LANL Chromium site. The ZEM framework will be released as open-source code and can be applied to any environmental management site. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

GIGA-Lens: Fast Bayesian Inference for Strong Gravitational Lens Modeling

We present GIGA-Lens: a gradient-informed, GPU-accelerated Bayesian framework for modeling strong gravitational lensing systems, implemented in TensorFlow and JAX. The three components, optimization using multistart gradient descent, posterior covariance estimation with variational inference, and sampling via Hamiltonian Monte Carlo, all take advantage of gradient information through automatic differentiation and massive parallelization on graphics processing units (GPUs). We test our pipeline on a large set of simulated systems and demonstrate in detail its high level of performance. The average time to model a single system on four Nvidia A100 GPUs is 105 s. The robustness, speed, and scalability offered by this framework make it possible to model the large number of strong lenses found in current surveys and present a very promising prospect for the modeling of ${ \mathcal O }({10}^{5})$ lensing systems expected to be discovered in the era of the Vera C. Rubin Observatory, Euclid, and the Nancy Grace Roman Space Telescope.

79 ASTRONOMY AND ASTROPHYSICS↗

Extending Parsimonious Bayesian Inference

Parsimonious Bayesian inference is a theoretical framework for efficient data assimilation that seeks to balance increased consistency between predictions and training data against corresponding increases in model complexity. Within this framework, over-training is understood as optimization that encodes excessive information within model parameters while only achieving small improvements between predictions and training data. This project aims to develop practical methods of limiting excess model information during optimization. One key observation is that practical heuristics for parsimonious learning in high-dimensions must balance expressivity, i.e. the ability of the model to capture diverse predictions with only a few non-zero parameters, against discoverability, i.e. the ability to train the model with gradient-based optimization and drive parameters to low information states. As such, we developed logical activation functions that are able to adaptively approximate arbitrary truth tables that define Boolean logic operations within a probabilistic framework. These functions have demonstrated the ability to learn exclusive disjunction (XOR) and conditioned disjunction (if [condition] then [result_if_true] else [result_if_false]) within a single layer of a neural network. To efficiently exploit these activation functions to drive parsimonious learning required several other advances within the domain of variational inference. The most efficient form of complexity suppression is structured sparsification, driving most model parameters to zero while achieving the structural coherence among nonzeros needed for bandwidth reduction. Such models are not only far more efficient at suppressing information-theoretic complexity, they also reduce the other forms of complexity (computations, communication, storage, and the number of dependencies needed to evaluate predictions). Aiming to support enhanced sparsification, this project examined new approaches to high-dimensional variational inference that allow us to calibrate and control parameter uncertainty during optimization. By identifying which parameters can sustain sparsifying perturbations with little impact on prediction quality, we can develop better pruning strategies by framing them as approximate Bayesian inference. These advances also open paths to mitigate concerns with deploying advanced learning methods in resource-constrained environments, such as running models on power-limited or communication-limited devices.

97 MATHEMATICS AND COMPUTING↗

Using Bayesian Inference Framework towards Identifying Gas Species and Concentration from High Temperature Resistive Sensor Array Data

High temperature gas sensors have been highly demanded for combustion process optimization and toxic emissions control, which usually suffer from poor selectivity. In order to solve this selectivity issue and identify unknown reducing gas species (CO, CH 4 , and CH 8 ) and concentrations, a high temperature resistive sensor array data set was built in this study based on 5 reported sensors. As each sensor showed specific responses towards different types of reducing gas with certain concentrations, based on which calibration curves were fitted, providing benchmark sensor array response database, then Bayesian inference framework was utilized to process the sensor array data and build a sample selection program to simultaneously identify gas species and concentration, by formulating proper likelihood between input measured sensor array response pattern of an unknown gas and each sampled sensor array response pattern in benchmark database. This algorithm shows good robustness which can accurately identify gas species and predict gas concentration with a small error of less than 10% based on limited amount of experiment data. These features indicate that Bayesian probabilistic approach is a simple and efficient way to process sensor array data, which can significantly reduce the required computational overhead and training data.

Liu, Yixin↗

Understanding the Impacts of Biomass Blending on the Uncertainty of Hydrolyzed Sugar Yield from a Stochastic Perspective

Feedstock price and availability are key challenges for biorefinery development and biomass blending has been suggested as a route to overcome these limitations. However, the impacts of feedstock blending on the variability of yields of hydrolyzed sugar remained unclear. This study quantifies the uncertainties in the sugar yields from the blends of corn stover, switchgrass, and grass clippings by considering both variable and model uncertainties. The results indicate that feedstock blending reduces the uncertainties in sugar yields and delivers feedstock of more uniform quality. A further investigation on xxx demonstrates that the reduction in overall uncertainty is achieved by reduced compositional uncertainties of blended feedstock. Based on these results, the optimization of blending ratios was performed with respect to three objectives: (1) probability of meeting the sugar yield target; (2) expected sugar yields; and (3) sugar yield per unit feedstock expense, while satisfying constraints of feedstock availability and cost. In conclusion, the optimization method developed in this study, is readily applied to other combinations of feedstocks, biofuels production processes, and constraints.

09 BIOMASS FUELS↗

Gamma-ray burst flares: X-ray flaring. II

We present a catalog of 498 flaring periods found in gamma-ray burst (GRB) light curves taken from the online Swift X-Ray Telescope GRB Catalogue. We analyzed 680 individual light curves using a flare detection method developed and used on our UV/optical GRB Flare Catalog. This method makes use of the Bayesian Information Criterion to analyze the residuals of fitted GRB light curves and statistically determines the optimal fit to the light curve residuals in an attempt to identify any additional features. These features, which we classify as flares, are identified by iteratively adding additional 'breaks' to the light curve. We find evidence of flaring in 326 of the analyzed light curves. For those light curves with flares, we find an average number of ∼1.5 flares per GRB. As with the UV/optical, flaring in our sample is generally confined to the first 1000 s of the afterglow, but can be detected to beyond 10{sup 5} s. Only ∼50% of the detected flares follow the 'classical' definition of Δt/t ≤ 0.5, with many of the largest flares exceeding this value.

79 ASTRONOMY AND ASTROPHYSICS↗

Statistical Treatment of Inverse Problems Constrained by Differential Equations-based Models with Stochastic Terms

This paper introduces a statistical treatment of inverse problems constrained by models with stochastic terms. The solution of the forward problem is given by a distribution represented numerically by an ensemble of simulations. The goal is to formulate the inverse problem, in particular the objective function, to find the closest forward distribution (i.e., the output of the stochastic forward problem) that best explains the distribution of the observations in a certain metric. We use proper scoring rules, a concept employed in statistical forecast verification, namely energy, variogram, and hybrid (i.e., combination of the two) scores. We study the performance of the proposed formulation in the context of two applications: a coefficient field inversion for subsurface flow governed by an elliptic partial differential equation with a stochastic source and a parameter inversion for power grid governed by differential-algebraic equations. In both cases we show that the variogram and the hybrid scores produce better parameter inversion results than does the energy score, whereas the energy score leads to better probabilistic predictions.

97 MATHEMATICS AND COMPUTING↗

Advanced Computational Framework for Environmental Management ZEM, Version 1.x

Typically environmental management problems require analysis of large and complex data sets originating from concurrent data streams with different data collection frequencies and pedigree. These big data sets require on-the-fly integration into a series of models with different complexity for various types of model analyses where the data are applied as soft and hard model constraints. This is needed to provide fast iterative model analyses based on the latest available data to guide decision-making. Furthermore, the data and model are associated with uncertainties. The uncertainties are probabilistic (e.g. measurement errors) and non-probabilistic (unknowns, e.g. alternative conceptual models characterizing site conditions). To address all of these issues, we have developed an integrated framework for real-time data and model analyses for environmental decision-making called ZEM. The framework allows for seamless and on-the-fly integration of data and modeling results for robust and scientifically-defensible decision-making applying advanced decision analyses tools such as Bayesian- Information-Gap Decision Theory (BIG-DT). The framework also includes advanced methods for optimization that are capable of dealing with a large number of unknown model parameters, and surrogate (reduced order) modeling capabilities based on support vector regression techniques. The framework is coded in Julia, a state-of-the-art high-performance programing language (http://julialang.org). The ZEM framework is open-source and can be applied to any environmental management site. The framework will be open-source and released under GPL V3 license.

Vesselinov, VelimirV.↗

Quantifying and reducing model-form uncertainties in Reynolds-averaged Navier–Stokes simulations: A data-driven, physics-informed Bayesian approach

Despite their well-known limitations, Reynolds-Averaged Navier–Stokes (RANS) models are still the workhorse tools for turbulent flow simulations in today's engineering analysis, design and optimization. While the predictive capability of RANS models depends on many factors, for many practical flows the turbulence models are by far the largest source of uncertainty. As RANS models are used in the design and safety evaluation of many mission-critical systems such as airplanes and nuclear power plants, quantifying their model-form uncertainties has significant implications in enabling risk-informed decision-making. In this work we develop a data-driven, physics-informed Bayesian framework for quantifying model-form uncertainties in RANS simulations. Uncertainties are introduced directly to the Reynolds stresses and are represented with compact parameterization accounting for empirical prior knowledge and physical constraints (e.g., realizability, smoothness, and symmetry). An iterative ensemble Kalman method is used to assimilate the prior knowledge and observation data in a Bayesian framework, and to propagate them to posterior distributions of velocities and other Quantities of Interest (QoIs). We use two representative cases, the flow over periodic hills and the flow in a square duct, to evaluate the performance of the proposed framework. Both cases are challenging for standard RANS turbulence models. Simulation results suggest that, even with very sparse observations, the obtained posterior mean velocities and other QoIs have significantly better agreement with the benchmark data compared to the baseline results. At most locations the posterior distribution adequately captures the true model error within the developed model form uncertainty bounds. The framework is a major improvement over existing black-box, physics-neutral methods for model-form uncertainty quantification, where prior knowledge and details of the models are not exploited. This approach has potential implications in many fields in which the governing equations are well understood but the model uncertainty comes from unresolved physical processes. - Highlights: • Proposed a physics–informed framework to quantify uncertainty in RANS simulations. • Framework incorporates physical prior knowledge and observation data. • Based on a rigorous Bayesian framework yet fully utilizes physical model. • Applicable for many complex physical systems beyond turbulent flows.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Conjunctive management of surface and groundwater resources under projected future climate change scenarios

Our study introduces a mixed integer linear fractional programming (MILFP) method to optimize conjunctive use of future surface water and groundwater resources under projected climate change scenarios. The conjunctive management model maximizes the ratio of groundwater usage to reservoir water usage. Future inflows to the reservoirs were estimated from the future runoffs projected through hydroclimate modeling considering the Variable Infiltration Capacity model, and 11 sets of downscaled Coupled Model Intercomparison Project phase 5 global climate model projections. Bayesian model averaging was adopted to quantify uncertainty in future runoff projections and reservoir inflow projections due to uncertain future climate projections. Optimized conjunctive management solutions were investigated for a water supply network in northern Louisiana which includes the Sparta aquifer. Runoff projections under climate change scenarios indicate that runoff will likely decrease in winter and increase in other seasons. Ultimately, results from the developed conjunctive management model with MILFP indicate that the future reservoir water, even at 2.5% low inflow cumulative probability level, could counterbalance groundwater pumping reduction to satisfy demands while improving the Sparta aquifer through conditional groundwater head constraint.

54 ENVIRONMENTAL SCIENCES↗