Search NASA⌕ Search

SEARCH · Search NASA

Results for “Deep Space Exploration”

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 127 records · Page 7
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

High-energy synchrotron flares powered by strongly radiative relativistic magnetic reconnection: 2D and 3D PIC simulations

ABSTRACT The time evolution of high-energy synchrotron radiation generated in a relativistic pair plasma energized by reconnection of strong magnetic fields is investigated with 2D and 3D particle-in-cell (PIC) simulations. The simulations in this 2D/3D comparison study are conducted with the radiative PIC code OSIRIS, which self-consistently accounts for the synchrotron radiation reaction on the emitting particles, and enables us to explore the effects of synchrotron cooling. Magnetic reconnection causes compression of the plasma and magnetic field deep inside magnetic islands (plasmoids), leading to an enhancement of the flaring emission, which may help explain some astrophysical gamma-ray flare observations. Although radiative cooling weakens the emission from plasmoid cores, it facilitates additional compression there, further amplifying the magnetic field B and plasma density n, and thus partially mitigating this effect. Novel simulation diagnostics utilizing 2D histograms in the n-B space are developed and used to visualize and quantify the effects of compression. The n-B histograms are observed to be bounded by relatively sharp power-law boundaries marking clear limits on compression. Theoretical explanations for some of these compression limits are developed, rooted in radiative resistivity or 3D kinking instabilities. Systematic parameter-space studies with respect to guide magnetic field, system size, and upstream magnetization are conducted and suggest that stronger compression, brighter high-energy radiation, and perhaps significant quantum electrodynamic effects such as pair production, may occur in environments with larger reconnection-region sizes and higher magnetization, particularly when magnetic field strengths approach the critical (Schwinger) field, as found in magnetar magnetospheres.

Schoeffler, K. M. (ORCID:0000000276447381)↗
DOE OSTI2023

Type Ia supernova observations combining data from the Euclid mission and the Vera C. Rubin Observatory

ABSTRACT The Euclid mission will provide first-of-its-kind coverage in the near-infrared over deep (three fields, ∼10–20 square degrees each) and wide (∼10 000 square degrees) fields. While the survey is not designed to discover transients, the deep fields will have repeated observations over a two-week span, followed by a gap of roughly six months. In this analysis, we explore how useful the deep field observations will be for measuring properties of Type Ia supernovae (SNe Ia). Using simulations that include Euclid’s planned depth, area, and cadence in the deep fields, we calculate that more than 3700 SNe between 0.0 < $z$ < 1.5 will have at least five Euclid detections around peak with signal-to-noise ratio larger than 3. While on their own, Euclid light curves are not good enough to directly constrain distances, when combined with legacy survey of space and time (LSST) deep field observations, we find that uncertainties on SN distances are reduced by 20–30 per cent for $z$ < 0.8 and by 40–50 per cent for $z$ > 0.8. Furthermore, we predict how well additional Euclid mock data can be used to constrain a key systematic in SN Ia studies – the size of the luminosity ‘step’ found between SNe hosted in high-mass (>1010 M⊙) and low-mass (<1010 M⊙) galaxies. This measurement has unique information in the rest-frame near-infrared (NIR). We predict that if the step is caused by dust, we will be able to measure its reduction in the NIR compared to optical at the 4σ level. We highlight that the LSST and Euclid observing strategies used in this work are still provisional and some level of joint processing is required. Still, these first results are promising, and assuming that Euclid begins observations well before the Nancy Roman Space Telescope (Roman), we expect this data set to be extremely helpful for preparation for Roman itself.

Astronomy & Astrophysics↗
DOE OSTI2022

FutureTense

Protective vaccines and reliable diagnostics are essential tools for controlling viral diseases. However, the efficacy of these tools can be diminished by mutations in viral genomes. The delay between the emergence of new viral strains and the redesign of vaccines and diagnostics allows for continued viral transmission. Is it possible to address this challenge by computationally predicting viral genome sequence evolution? Can we “future-proof” vaccines and diagnostics by targeting both current and anticipated future sequence variants? While predicting viral evolution is still an unsolved, “grand challenge” problem in biology, the large, and rapidly growing, number of SARS-CoV-2 genome sequences provide an opportunity to quantify the ability of machine learning to predict viral genome sequence evolution. Towards this end, we have developed a simple computational model for predicting viral evolution at the level of individual nucleotides. The key metric for quantifying the per-base, prediction accuracy for viral evolution is the Mann-Whitney U statistic (or, equivalently, the area under the receiver operator curve). Since the Mann-Whitney U statistic is not a differentiable function, existing deep leaning packages (like Pytorch and Keras/TensorFlow) are not useful, as they require that the accuracy metric/objective function be analytically differentiable with respect to the model parameters. To overcome this challenge, we have implemented custom software, “FutureTense”, that can train a machine learning model by maximizing the non-differentiable Mann-Whitney U statistic. This software trains a machine learning model by exploring along the direction of the discrete gradient of the Mann-Whitney U statistic in the model parameter space. Parallel computing and genome sequence-specific optimizations are used to accelerate model training. The resulting machine learning model learns the observed high C->U mutation rates in the SARS-CoV-2 genome (which are potentially induced by host defenses) and provides prediction accuracies that are significantly better than one would expect from random chance. While predicting viral evolution is still quite far from a solved problem, the surprising performance of this simple model gives hope that the accuracy of predicting viral genome evolution can be further increased by more sophisticated approaches.

Gans, Jason↗
DOE OSTI2025

Space environment considerations for perovskite solar cell operations: A review

Designing new technology for extraterrestrial applications certainly presents unique challenges. The environmental stressors perovskite-based photovoltaics must overcome will vary with the environment in which they are deployed. One must consider mission requirements when designing photovoltaic devices and packaging. Different space "theaters" can have dramatically different stressors needing consideration for designing panels for solar power generation. Here, in this article, we review the relevant space environmental conditions that must be considered when designing perovskite-based photovoltaic devices for implementation in space. We specifically consider thermal, radiation, gaseous, weather, and other phenomena most relevant to photovoltaic operation for specific theaters such as Low Earth Orbit, Geosynchronous Orbit, Lunar surface, Mars (orbit and surface), and interplanetary exploration pathways.

14 SOLAR ENERGY↗
DOE OSTI2023

GAHLS: an optimized graph analytics based high level synthesis framework

The urgent need for low latency, high-compute and low power on-board intelligence in autonomous systems, cyber-physical systems, robotics, edge computing, evolvable computing, and complex data science calls for determining the optimal amount and type of specialized hardware together with reconfigurability capabilities. With these goals in mind, we propose a novel comprehensive graph analytics based high level synthesis (GAHLS) framework that efficiently analyzes complex high level programs through a combined compiler-based approach and graph theoretic optimization and synthesizes them into message passing domain-specific accelerators. This GAHLS framework first constructs a compiler-assisted dependency graph (CaDG) from low level virtual machine (LLVM) intermediate representation (IR) of high level programs and converts it into a hardware friendly description representation. Next, the GAHLS framework performs a memory design space exploration while account for the identified computational properties from the CaDG and optimizing the system performance for higher bandwidth. The GAHLS framework also performs a robust optimization to identify the CaDG subgraphs with similar computational structures and aggregate them into intelligent processing clusters in order to optimize the usage of underlying hardware resources. Finally, the GAHLS framework synthesizes this compressed specialized CaDG into processing elements while optimizing the system performance and area metrics. Evaluations of the GAHLS framework on several real-life applications (e.g., deep learning, brain machine interfaces) demonstrate that it provides 14.27× performance improvements compared to state-of-the-art approaches such as LegUp 6.2.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Exploring the Nucleon Structure via Deep Electroproduction Processes

Understanding the internal structure of the nucleon is a fundamental goal of modern physics, which aims at a comprehensive framework describing the internal dynamics of quarks and gluons. Among other structure functions, Generalized Parton Distributions (GPDs) offer a powerful framework for describing the nucleon dynamics by correlating the longitudinal momentum and the transverse position of its internal partons. Such a correlation provides a three-dimensional picture of the nucleon and enables access to fundamental properties, including the internal pressure distributions and the parton’s angular momentum contribution to the nucleon's total spin, thereby playing a central role in resolving the nucleon spin puzzle. At the Thomas Jefferson National Facility (JLab), polarized electron beam experiments allow for probing GPDs through the measurement of hard exclusive processes. Among the cleanest experimental channels, we find the electro-production of a real photon through the Deeply Virtual Compton Scattering (DVCS) mechanism. The first data-taking period of the CLAS12 program, taking place in 2018, allowed for unique DVCS Beam Spin Asymmetry (BSA) measurements in the phase space covered by a 10.6 GeV polarized electron beam impinging on an unpolarized liquid hydrogen target. Although detecting all final-state particles ensures exclusivity of the process, conservation laws indicate that it is not mandatory. I adopt an approach omitting the direct detection of the recoil proton, providing a simplified yet effective event selection strategy that boosts statistics and gives access to a larger phase space sensitive to the underlying GPD dynamics through BSA and cross section measurements. The Double DVCS (DDVCS) process promises a dedicated mapping of GPDs. Contributing to the electro-production of a lepton pair cross-section, the DDVCS reaction extends DVCS by allowing the final-state photon to be virtual, enriching the kinematic phase space and providing unique access to the internal correlations encoded by GPDs. A feasibility study is conducted to assess the potential of future DDVCS measurements at Jefferson Lab and the future Electron-Ion Collider (EIC). While Jefferson Lab will provide DDVCS measurements in the valence region through the SoLID$\mu$ and $\mu$CLAS12 experimental projects, in the long term, the EIC will provide complementary measurements in the sea region, both accessing unprecedented information about GPDs in a phase space region otherwise inaccessible. Taken together, these investigations demonstrate both the current capabilities and future opportunities for probing GPDs through exclusive processes. The experimental analysis of DVCS at CLAS12 provides precise measurements within an established framework, while the phenomenological study of DDVCS opens the door to richer and more comprehensive explorations with future detectors and facilities.

Alvarado, Juan [Université Paris-Saclay: Gif-sur-Y↗
DOE OSTI2021

Deep Reinforcement Scheduling of Energy Storage Systems for Real-time Voltage Regulation in Unbalanced LV Networks with High PV Penetration

The ever-growing higher penetration of distributed energy resources (DERs) in low-voltage (LV) distribution systems brings both opportunities and challenges to voltage support and regulation. This paper proposes a deep reinforcement learning (DRL)-based scheduling scheme of energy storage systems (ESSs) to mitigate system voltage deviations in unbalanced LV distribution networks. The ESS-based voltage regulation problem is formulated as a multi-stage quadratic stochastic program, with the objective of minimizing the expected total daily voltage regulation cost while satisfying operational constraints. While existing voltage regulation methods are mostly focused on onetime- step control, this paper explores a day-horizon systemwide voltage regulation problem. In other words, the size of action and state spaces are extremely high-dimensional and need to be delicately handled. Furthermore, in order to overcome the difficulty of modeling uncertainties and develop a realtime solution, a learn-to-schedule feedback control framework is proposed by adapting the problem to a model-free DRL setting. The proposed algorithm is tested on a customized 6-bus system and a modified IEEE 34-bus system. Simulation results validate the effectiveness and near-optimality of voltage regulation by ESS in comparison with a deterministic quadratic program solution.

Wang, Shengyi↗
DOE OSTI2022

ADDGALS: Simulated Sky Catalogs for Wide Field Galaxy Surveys

Abstract We present a method for creating simulated galaxy catalogs with realistic galaxy luminosities, broadband colors, and projected clustering over large cosmic volumes. The technique, denoted Addgals (Adding Density Dependent GAlaxies to Lightcone Simulations), uses an empirical approach to place galaxies within lightcone outputs of cosmological simulations. It can be applied to significantly lower-resolution simulations than those required for commonly used methods such as halo occupation distributions, subhalo abundance matching, and semi-analytic models, while still accurately reproducing projected galaxy clustering statistics down to scales of r ∼ 100 h −1 kpc . We show that Addgals catalogs reproduce several statistical properties of the galaxy distribution as measured by the Sloan Digital Sky Survey (SDSS) main galaxy sample, including galaxy number densities, observed magnitude and color distributions, as well as luminosity- and color-dependent clustering. We also compare to cluster–galaxy cross correlations, where we find significant discrepancies with measurements from SDSS that are likely linked to artificial subhalo disruption in the simulations. Applications of this model to simulations of deep wide-area photometric surveys, including modeling weak-lensing statistics, photometric redshifts, and galaxy cluster finding, are presented in DeRose et al., and an application to a full cosmology analysis of Dark Energy Survey (DES) Year 3 like data is presented in DeRose et al. We plan to publicly release a 10,313 square degree catalog constructed using Addgals with magnitudes appropriate for several existing and planned surveys, including SDSS, DES, VISTA, Wide-field Infrared Survey Explorer, and Rubin Observatory’s Legacy Survey of Space and Time.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2021

Semi-supervised Bayesian Low-shot Learning

Deep neural networks (NNs) typically outperform traditional machine learning (ML) approaches for complicated, non-linear tasks. It is expected that deep learning (DL) should offer superior performance for the important non-proliferation task of predicting explosive device configuration based upon observed optical signature, a task which human experts struggle with. However, supervised machine learning is difficult to apply in this mission space because most recorded signatures are not associated with the corresponding device description, or “truth labels.” This is challenging for NNs, which traditionally require many samples for strong performance. Semi-supervised learning (SSL), low-shot learning (LSL), and uncertainty quantification (UQ) for NNs are emerging approaches that could bridge the mission gaps of few labels and rare samples of importance. NN explainability techniques are important in gaining insight into the inferential feature importance of such a complex model. In this work, SSL, LSL, and UQ are merged into a single framework, a significant technical hurdle not previously demonstrated. Exponential Average Adversarial Training (EAAT) and Pairwise Neural Networks (PNNs) are chosen as the SSL and LSL methods of choice. Permutation feature importance (PFI) for functional data is used to provide explainability via the Variable importance Explainable Elastic Shape Analysis (VEESA) pipeline. A variety of uncertainty quantification approaches are explored: Bayesian Neural Networks (BNNs), ensemble methods, concrete dropout, and evidential deep learning. Two final approaches, one utilizing ensemble methods and one utilizing evidential learning, are constructed and compared using a well-quantified synthetic 2D dataset along with the DIRSIG Megascene.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Deep Gaussian process-based cost-aware batch Bayesian optimization for complex materials design campaigns

The accelerating pace and expanding scope of materials discovery demand optimization frameworks that efficiently navigate vast design spaces with complex response surfaces while judiciously allocating limited evaluation resources. We present a cost-aware, batch Bayesian optimization scheme powered by deep Gaussian process (DGP) surrogates and a heterotopic querying strategy. Our DGP surrogate, formed by stacking GP layers, models complex hierarchical relationships among high-dimensional compositional features and captures correlations across multiple target properties, propagating uncertainty through successive layers. We integrate evaluation cost into an upper-confidence-bound acquisition extension, which, together with heterotopic querying, proposes small batches of candidates in parallel, balancing exploration of under-characterized regions with exploitation of high-mean, low-variance predictions across correlated properties. Applied to refractory high-entropy alloys for high-temperature applications, our framework converges to optimal formulations in fewer iterations with cost-aware queries than conventional GP-based BO, highlighting the value of deep, uncertainty-aware, cost-sensitive strategies in materials campaigns.

36 MATERIALS SCIENCE↗
DOE OSTI2020

Mars Extant Life: What's Next? Conference Report

On November 5–8, 2019, the ‘‘Mars Extant Life: What’s Next?’’ conference was convened in Carlsbad, New Mexico. The conference gathered a community of actively publishing experts in disciplines related to habitability and astrobiology. Primary conclusions are as follows: A significant subset of conference attendees concluded that there is a realistic possibility that Mars hosts indigenous microbial life. A powerful theme that permeated the conference is that the key to the search for martian extant life lies in identifying and exploring refugia (‘‘oases’’), where conditions are either permanently or episodically significantly more hospitable than average. Based on our existing knowledge of Mars, conference participants highlighted four potential martian refugium (not listed in priority order): Caves, Deep Subsurface, Ices, and Salts. The conference group did not attempt to reach a consensus prioritization of these candidate environments, but instead felt that a defensible prioritization would require a future competitive process. Within the context of these candidate environments, we identified a variety of geological search strategies that could narrow the search space. Additionally, we summarized a number of measurement techniques that could be used to detect evidence of extant life (if present). Again, it was not within the scope of the conference to prioritize these measurement techniques—that is best left for the competitive process. We specifically note that the number and sensitivity of detection methods that could be implemented if samples were returned to Earth greatly exceed the methodologies that could be used at Mars. Finally, important lessons to guide extant life search processes can be derived both from experiments carried out in terrestrial laboratories and analog field sites and from theoretical modeling.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2026

Accelerating Traction Motor Optimization Design with AI Surrogate Models

The advancement of artificial intelligence systems enables the use of data-driven physics-based surrogate models to explore design spaces rapidly and deeply for engineering projects. This work presents a surrogate model workflow that accelerates electric traction motor design optimization by replacing finite element analysis (FEA) with an artificial neural network (ANN) and using this model in a genetic algorithm for design optimization. A baseline interior permanent-magnet motor is parameterized and sampled to generate FEA-labeled training data, after which a feed-forward ANN predicts key outputs (e.g., loss components and weight). The validated surrogate enables genetic-algorithm optimization and deep search over the design space without new FEA runs, producing Pareto-optimal trade-offs between weight and losses and set of optimized designs for rapid downselection of manufacturable motor designs.

Ribeiro, Pedro [ORNL] (ORCID:0009000921026641)↗
DOE OSTI2024

Graph neural networks for CO 2 solubility predictions in Deep Eutectic Solvents

Deep Eutectic Solvents (DESs) are a promising class of solvents for CO 2 capture. DESs are complex mixtures that can be designed to optimize CO solubility and overall capture process efficiency. However, the vast design landscape of DES mixtures makes experimental investigation prohibitive; as such, there is a need for computational models that can quickly and efficiently navigate the design space and inform data collection efforts. In this work, we propose Graph Neural Network (GNN) models for predicting CO 2 solubility for DESs; the GNN leverages a mixture graph representation that captures the molecular structure of the DES components as well as their intermolecular interactions. Here, we compare the GNN framework against alternative architectures (neural networks, graph convolution networks, and random forests) and data representations (molecular fingerprints, sigma profiles, and graphs). We show that the proposed approach offers superior predictive performance; specifically, we show that solubility can be predicted reliably directly from molecular structure (without the need of using sigma profiles as proposed in previous studies). This result is important, as obtaining sigma profiles requires expensive density functional theory computations. We also explored the ability of GNNs to predict solubility for new DES mixtures and operating conditions. We found that the model extrapolates across temperature reliably. However, we also found deficiencies in the ability of the model to predict solubility for DES mixtures, pressures, and molar ratio not included in the training sets; we show that this is due to an inherent lack of chemical diversity in datasets available in the literature. The proposed computational capabilities can thus help navigate the design space of DES and inform data collection efforts. Our models, data, and benchmarks are shared as Python code implemented in Jupyter notebooks.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2023

DEEPEN Leapfrog Geodata Model Cleaned and Reformatted Exploration Datasets from Newberry Volcano

DEEPEN stands for DE-risking Exploration of geothermal Plays in magmatic ENvironments. As part of the DEEPEN 3D play fairway analysis (PFA) conducted at Newberry Volcano for multiple play types (conventional hydrothermal, superhot EGS, and supercritical), existing geoscientific exploration datasets needed to be acquired, cleaned, reformatted, and assembled in Leapfrog Geothermal. This GDR submission includes all of the cleaned and reformatted (X (m), Y (m), elevation (m), processed data values) datasets used to build the Leapfrog Geodata model. Existing datasets were acquired from the GDR, from AltaRock, and from other sources. This yielded the following datasets: - Digital elevation model produced from LiDAR data by Ramsey and Bard, 2016 - MT surveys from 2006, 2011, 2014, and 2017 (including single inversions) - Gravity surveys from 2006, 2007, and 2011 (including single) - Earthquake catalogs from PNSN, LLNL, and the Newberry EGS Demonstration project - Seismic velocity model from Templeton et al., 2014 - The Frone, 2015 temperature model and a new one produced through extrapolating downhole temperature measurements and the SMU temperature at depth maps. Two versions of the new model are provided: 250 m spacing and 500 m spacing - EarthVision geologic model with alteration from Moser et al., 2016 - Well data from EGS well 55-29, deep geothermal wells, coreholes (GEO N-2 through 5) and several thermal gradient holes - "Newberry Well Data:" Location, simple lithology, directional survey data, and temperature data for the 34 wells and coreholes used in the Newberry PFA Although there are additional 2D datasets available in the area, such as aeromagnetic surveys, these were not included in the analysis. While it may be possible to project these datasets into three dimensions by assuming the surface measurements do not vary with depth, this method is associated with high uncertainty. Preexisting inversions of these data were unavailable, and inverting additional geophysical datasets is outside the scope of this project.

15 GEOTHERMAL ENERGY↗
DOE OSTI2022

Idealized Simulations of the Tropical Climate and Variability in the Single Column Atmosphere Model (SCAM): Radiative-Convective Equilibrium

To explore the interactions among column processes in the Community Atmosphere Model (CAM), the single-column version of CAM (SCAM) is integrated for 1000 days in radiative-convective equilibrium (RCE) with tropical values of boundary conditions, spanning a parameter or configuration space of model physics versions (v5 vs. v6), vertical resolution (standard and 60 levels), sea surface temperature (SST), and some interpretation-driven experiments. The simulated time-mean climate is reasonable, near observations and RCE of a cyclic cloud-resolving model. Updraft detrainment in the deep convection scheme produces distinctive grid-scale structures in humidity and cloud, which also interact with radiative transfer processes. These grid artifacts average out in multi-column RCE results reported elsewhere, illustrating the nuts-and-bolts interpretability that SCAM adds to the hierarchy of model configurations. Multi-day oscillations of precipitation arise from descent of warm convection-capping layers starting near the tropopause, eventually reset by a burst of convective deepening. Experiments reveal how these oscillations depend critically on an internal parameter that controls the number of neutral buoyancy levels allowed for determining cloud top and computing dilute convective available potential energy in the deep convection scheme, and merely modified a little by disabling cloud-base radiation (heating of cloud base). This strong dependence of transient behavior in 1D on this parameter will be tested in the second part of this work, in which SCAM is coupled to a parameterized dynamics of two-dimensional, linearized gravity wave, and in the 3D simulations in future study.

Meteorology & Atmospheric Sciences↗
DOE OSTI2022

Deep Realistic Extragalactic Model (DREaM) Galaxy Catalogs: Predictions for a Roman Ultra-deep Field

In the next decade, deep galaxy surveys from telescopes such as the James Webb Space Telescope and Roman Space Telescope will provide transformational data sets that will greatly enhance the understanding of galaxy formation during the epoch of reionization (EoR). In this work, we present the Deep Realistic Extragalactic Model (DREaM) for creating synthetic galaxy catalogs. Our model combines dark matter simulations, subhalo abundance matching and empirical models, and includes galaxy positions, morphologies, and spectral energy distributions. The resulting synthetic catalog extends to redshifts z ~ 12, and galaxy masses log 10 (M/M ⊙ ) = 5 covering an area of 1 deg 2 on the sky. We use DREaM to explore the science returns of a 1 deg 2 Roman ultra-deep field (UDF), and to provide a resource for optimizing ultra-deep survey designs. We find that a Roman UDF to ~30 m AB will potentially detect more than 10 6 M UV < -17 galaxies, with more than 10 4 at redshifts z > 7, offering an unparalleled data set for constraining galaxy properties during the EoR. Our synthetic catalogs and simulated images are made publicly available to provide the community with a tool to prepare for upcoming data.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2022

Learning to identify semi-visible jets

We train a network to identify jets with fractional dark decay (semi-visible jets) using the pattern of their low-level jet constituents, and explore the nature of the information used by the network by mapping it to a space of jet substructure observables. Semi-visible jets arise from dark matter particles which decay into a mixture of dark sector (invisible) and Standard Model (visible) particles. Such objects are challenging to identify due to the complex nature of jets and the alignment of the momentum imbalance from the dark particles with the jet axis, but such jets do not yet benefit from the construction of dedicated theoretically-motivated jet substructure observables. A deep network operating on jet constituents is used as a probe of the available information and indicates that classification power not captured by current high-level observables arises primarily from low-p T jet constituents.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2023

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↗