SEARCH · Search NASA

Results for “high performance and distributed computing”

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.

SourceResearch connectionFrom yearThrough year

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

At least 1,405 records · Page 78

All sources NASA NTRS DOE OSTI

DOE OSTI2024

Uncovering the three-dimensional structure of upconverting core–shell nanoparticles with multislice electron ptychography

In photon upconverting core–shell nanoparticles, structure strongly dictates performance. Typical imaging in scanning transmission electron microscopy has sufficient resolution to probe the atomic structure of these nanoparticles, but contrast, dose, and projection limitations make conventional methods insufficient for fully characterizing these structures. Phase retrieval techniques provide a promising alternative imaging mode, and, in particular, multislice electron ptychography can recover depth-dependent information. Here, we study beam-sensitive photon upconverting core–shell nanoparticles with a multislice ptychography approach using a low electron dose to avoid damage. Large strain fields arise in these heterostructures due to the mismatch in lattice parameter between the core and the shell. We reconstruct both a nanoparticle that appears defect-free and one that has a large break in the side and map the distribution of strain in 3D by computing distortion fields from high-resolution potential images of each slice. In the defect-free nanoparticle, we observe twisting of the shell, while in the broken nanoparticle, we measure the 3D position of the crack, the core, and dislocations. These results highlight the advantage of multislice electron ptychography to recover 3D information from a single scan, even under strict electron dose requirements from beam-sensitive samples.

74 ATOMIC AND MOLECULAR PHYSICS↗

DOE OSTI2024

Optimizing Distributed Training on Frontier for Large Language Models

Large language models (LLMs) have demonstrated remarkable success as foundational models, benefiting various downstream applications through fine-tuning. Loss scaling studies have demonstrated the superior performance of larger LLMs compared to their smaller counterparts. Nevertheless, training LLMs with billions of parameters poses significant challenges and requires considerable computational resources. For example, training a one trillion parameter GPT-style model on 20 trillion tokens requires a staggering 120 million exaflops. This research explores efficient distributed training strategies to extract this computation from Frontier, the world's first exascale supercomputer. We enable and investigate various model and data parallel training techniques, such as tensor parallelism, pipeline parallelism, and sharded data parallelism, to facilitate training a trillion-parameter model on Frontier. We empirically assess these techniques and their associated parameters to determine their impact on memory footprint, communication latency, and GPU's computational efficiency. We analyze the complex interplay among these techniques and find a strategy to combine them to achieve high throughput through hyperparameter tuning. We have identified efficient strategies for training large LLMs of varying sizes through empirical analysis and hyperparameter tuning. For 22 Billion, 175 Billion, and 1 Trillion parameters, we achieved GPU throughputs of 38.38%, 36.14%, and 31.96%, respectively. For the training of the 175 Billion parameter model and the 1 Trillion parameter model, we achieved 100% weak scaling efficiency on 1024 and 3072 Mi250X GPUs, respectively. We also achieved strong scaling efficiencies of 89% and 87% for these two models. We trained these models only tens of iterations instead of training till completion.

Yin, Junqi↗

NASA NTRS2000

Analysis of Post-Support and Wind-Tunnel Wall Interference on Flow Field About Subsonic High-Lift High-Speed Research Configuration

The present study was performed to determine how significant the interference effects of the wind-tunnel model support system and tunnel walls can be for a high-speed configuration during takeoff and landing conditions. A 5-percent scale model of the Technology Concept Airplane was recently tested in the Langley 14- by 22-Foot Sub-sonic Tunnel. The model was numerically modeled with and without the support and tunnel walls and compared with experimental data. Detailed analysis of the flow provided additional insight concerning what effects the post support and tunnel walls had on the flow field. This study revealed that although the overall forces and moments could be experimentally accounted for, the detailed flow features, such as the surface pressure distributions, could not be accurately simulated without including the post support in the computations.

Lessard, Wendy B.↗

NASA NTRS2012

The Gas Distribution in Galaxy Cluster Outer Regions

Aims. We present the analysis of a local (z = 0.04 - 0.2) sample of 31 galaxy clusters with the aim of measuring the density of the X-ray emitting gas in cluster outskirts. We compare our results with numerical simulations to set constraints on the azimuthal symmetry and gas clumping in the outer regions of galaxy clusters. Methods. We exploit the large field-of-view and low instrumental background of ROSAT/PSPC to trace the density of the intracluster gas out to the virial radius. We perform a stacking of the density profiles to detect a signal beyond r200 and measure the typical density and scatter in cluster outskirts. We also compute the azimuthal scatter of the profiles with respect to the mean value to look for deviations from spherical symmetry. Finally, we compare our average density and scatter profiles with the results of numerical simulations. Results. As opposed to some recent Suzaku results, and confirming previous evidence from ROSAT and Chandra, we observe a steepening of the density profiles beyond approximately r(sub 500). Comparing our density profiles with simulations, we find that non-radiative runs predict too steep density profiles, whereas runs including additional physics and/or treating gas clumping are in better agreement with the observed gas distribution. We report for the first time the high-confidence detection of a systematic difference between cool-core and non-cool core clusters beyond 0.3r(sub 200), which we explain by a different distribution of the gas in the two classes. Beyond r(sub 500), galaxy clusters deviate significantly from spherical symmetry, with only little differences between relaxed and disturbed systems. We find good agreement between the observed and predicted scatter profiles, but only when the 1% densest clumps are filtered out in the simulations. Conclusions. Comparing our results with numerical simulations, we find that non-radiative simulations fail to reproduce the gas distribution, even well outside cluster cores. Although their general behavior is in better agreement with the observations, simulations including cooling and star formation convert a large amount of gas into stars, which results in a low gas fraction with respect to the observations. Consequently, a detailed treatment of gas cooling, star formation, AGN feedback, and taking into account gas clumping is required to construct realistic models of cluster outer regions.

Eckert, D.↗

DOE OSTI2023

Pulse profile modelling of thermonuclear burst oscillations – II. Handling variability

ABSTRACT Pulse profile modelling is a relativistic ray-tracing technique that can be used to infer masses, radii, and geometric parameters of neutron stars. In a previous study, we looked at the performance of this technique when applied to thermonuclear burst oscillations from accreting neutron stars. That study showed that ignoring the variability associated with burst oscillation sources resulted in significant biases in the inferred mass and radius, particularly for the high count rates that are nominally required to obtain meaningful constraints. In this follow-on study, we show that the bias can be mitigated by slicing the bursts into shorter segments where variability can be neglected, and jointly fitting the segments. Using this approach, the systematic uncertainties on the mass and radius are brought within the range of the statistical uncertainty. With about 106 source counts, this yields uncertainties of approximately 10 per cent for both the mass and radius. However, this modelling strategy requires substantial computational resources. We also confirm that the posterior distributions of the mass and radius obtained from multiple bursts of the same source can be merged to produce outcomes comparable to that of a single burst with an equivalent total number of counts.

Astronomy & Astrophysics↗

DOE OSTI2018

Efficient Stochastic Inversion Using Adjoint Models and Kernel-PCA

Performing stochastic inversion on a computationally expensive forward simulation model with a high-dimensional uncertain parameter space (e.g. a spatial random field) is computationally prohibitive even with gradient information provided. Moreover, the ‘nonlinear’ mapping from parameters to observables generally gives rise to non-Gaussian posteriors even with Gaussian priors, thus hampering the use of efficient inversion algorithms designed for models with Gaussian assumptions. In this work, we propose a novel Bayesian stochastic inversion methodology, characterized by a tight coupling between a gradient-based Langevin Markov Chain Monte Carlo (LMCMC) method and a kernel principal component analysis (KPCA). This approach addresses the ‘curse-of-dimensionality’ via KPCA to identify a low-dimensional feature space within the high-dimensional and nonlinearly correlated spatial random field. Moreover, non-Gaussian full posterior probability distribution functions are estimated via an efficient LMCMC method on both the projected low-dimensional feature space and the recovered high-dimensional parameter space. We demonstrate this computational framework by integrating and adapting recent developments such as data-driven statistics-on-manifolds constructions and reduction-through-projection techniques to solve inverse problems in linear elasticity.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2016

CFD and X-ray Analysis of Gaseous Direct Injection from an Outward Opening Injector

Using natural gas in an internal combustion engine (ICE) is emerging as a promising way to improve thermal efficiency and reduce exhaust emissions. In the development of such engine platforms, computational fluid dynamics (CFD) plays a fundamental role in the optimization of geometries and operating parameters. One of the most relevant issues in the simulation of direct injection (DI) gaseous processes is the accurate prediction of the gas jet evolution. The simulation of the injection process for a gaseous fuel does not require complex modeling, nevertheless properly describing high-pressure gas jets remains a challenging task. At the exit of the nozzle, the injected gas is under-expanded, the flow becomes supersonic and shocks occur due to compressibility effects. These phenomena lead to challenging computational requirements resulting from high grid resolution and low computational time-steps. In this paper, the simulation of gaseous injection from an outward opening injector was carried out with the commercial CFD software CONVERGE. Argon was injected into a quiescent chamber filled with Nitrogen. Two free-jet cases with injection pressure of 15 and 10 bar, and ambient pressure of 1 and 2 bar respectively, were tested. For the 15 bar injection pressure and 1 bar ambient pressure, a jet impinging on a barrier case was performed as well. The numerical results were validated against experimental mass measurements using X-ray radiography. The results, presented in terms of 2D maps and transverse sections of equivalent path length of argon, showed good agreement between modeling and experiments in terms of jet penetration, shape and gas mass distribution. These emphasized the significant impact of mesh size on the results for the simulation of this type of injector. Due to the small geometrical dimensions a fine mesh is required in order to deliver an accurate description of the mixture formation. Numerical results were discussed with a particular emphasis on the application to engine simulations, where a trade-off between accuracy and computational time is a necessity.

Bartolucci, Lorenzo↗

NASA NTRS1999

Automatic Data Distribution for CFD Applications on Structured Grids

Development of HPF versions of NPB and ARC3D showed that HPF has potential to be a high level language for parallelization of CFD applications. The use of HPF requires an intimate knowledge of the applications and a detailed analysis of data affinity, data movement and data granularity. Since HPF hides data movement from the user even with this knowledge it is easy to overlook pieces of the code causing low performance of the application. In order to simplify and accelerate the task of developing HPF versions of existing CFD applications we have designed and partially implemented ADAPT (Automatic Data Distribution and Placement Tool). The ADAPT analyzes a CFD application working on a single structured grid and generates HPF TEMPLATE, (RE)DISTRIBUTION, ALIGNMENT and INDEPENDENT directives. The directives can be generated on the nest level, subroutine level, application level or inter application level. ADAPT is designed to annotate existing CFD FORTRAN application performing computations on single or multiple grids. On each grid the application can considered as a sequence of operators each applied to a set of variables defined in a particular grid domain. The operators can be classified as implicit, having data dependences, and explicit, without data dependences. In order to parallelize an explicit operator it is sufficient to create a template for the domain of the operator, align arrays used in the operator with the template, distribute the template, and declare the loops over the distributed dimensions as INDEPENDENT. In order to parallelize an implicit operator, the distribution of the operator's domain should be consistent with the operator's dependences. Any dependence between sections distributed on different processors would preclude parallelization if compiler does not have an ability to pipeline computations. If a data distribution is "orthogonal" to the dependences of an implicit operator then the loop which implements the operator can be declared as INDEPENDENT.

Frumkin, Michael↗

DOE OSTI2019

Simultaneous three-dimensional elemental mapping of Hollandite and Pyrochlore material phases in ceramic waste form materials

In this study, differential X-ray absorption contrast tomography was performed on model ceramic composite systems to image the 3D spatial distributions of Ga and Nd containing phases. The model material systems were fabricated such that they contained two constituent phases. Ga–doped Ba hollandite (Ba 1.33 Ga 2.66 Ti 5.34 O 16 ), and pyrochlore (Nd 2 Ti 2 O 7 ) which are under consideration as nuclear waste immobilization matrices. Two complementary techniques have been used to characterize the 3D distribution of phases. The analysis suggests that the 3D spatial distributions of hollandite and pyrochlore can be readily identified through the distributions Ga and Nd, respectively. These results represent a critical development towards characterizing the complex microstructure inherent to practical materials such as multiphase ceramic composite waste forms. Furthermore, these methods can be applied to elucidate microstructural features that fundamentally determine the performance and properties of complex systems, which is central to the design of durable waste forms as well as many other real materials such fuel cells, batteries, ultra-high temperature ceramics.

36 MATERIALS SCIENCE↗

DOE OSTI2019

Short- and Intermediate-Range Structure and Dynamics of Fe-Ni-C Liquid Under Compression

Properties of liquid Fe alloys under high-pressure conditions are crucial for understanding the composition, thermal state, and dynamics of Earth’s core. Experiments on such liquids, however, are often performed under pressures far below those of the outer core, necessitating long extrapolations of experimental results to core conditions. Such estimates can be complicated by light elements possibly forming pressure-dependent molecular clusters that can significantly affect the physical properties of liquids as core conditions are approached. First-principles molecular dynamics simulations were employed to compute the properties of an Fe-Ni-C liquid with a composition of Fe 3.7 Ni 0.37 C at 1673 K and pressures from 0 to 67 GPa to benchmark computational methods on pressure effects on the structure and properties of the liquid relative to low pressure experimental results. The short-range structure is manifested by the coordination number (CN) of Fe/Ni-Fe/Ni being around 12, indicative of a nearly close-packed structure in the pressure range, and the CN of C-Fe/Ni gradual increasing from 6.5 to 8.5, indicative of an approximately octahedral to cubic transition as pressure increases. The Fe/Ni-Fe/Ni bond distance, however, is found to be 10 times more compressible than the C-Fe/Ni distance. The intermediate-range structure of Fe/Ni-Fe/Ni and C-Fe/Ni subsystems, described by a partial configurationally decomposed distribution function, undergoes substantial changes, characterized by a significant increase of the number of polyhedra that share 3 atoms with each other. Such dense configurations are related to an increased bulk modulus, decreased diffusion coefficient, decreased activation volume for diffusion, and increased shear viscosity. Reproducing the experimental observations at low pressures provides important support for modeling the liquid under the conditions relevant to the outer core.

58 GEOSCIENCES↗

DOE OSTI2022

Resilience Enhancements through Deep Learning Yields

This report documents the Resilience Enhancements through Deep Learning Yields (REDLY) project, a three-year effort to improve electrical grid resilience by developing scalable methods for system operators to protect the grid against threats leading to interrupted service or physical damage. The computational complexity and uncertain nature of current real-world contingency analysis presents significant barriers to automated, real-time monitoring. While there has been a significant push to explore the use of accurate, high-performance machine learning (ML) model surrogates to address this gap, their reliability is unclear when deployed in high-consequence applications such as power grid systems. Contemporary optimization techniques used to validate surrogate performance can exploit ML model prediction errors, which necessitates the verification of worst-case performance for the models.

24 POWER TRANSMISSION AND DISTRIBUTION↗

DOE OSTI2025

HARD: A performance portable radiation hydrodynamics code based on FleCSI framework

Hydrodynamics And Radiation Diffusion (HARD) is an open-source application for high-performance simulations of compressible hydrodynamics with radiation-diffusion coupling. Built on the FleCSI (Bergen et al., 2021 [1]) (Flexible Computational Science Infrastructure) framework, HARD expresses its computational units as tasks whose execution can be orchestrated by multiple back-end runtimes, including Legion (Bauer et al., 2012 [2]), MPI (Forum, 1994 [3]), and HPX (Kaiser et al., 2020 [4]). Node-level parallelism is handled through Kokkos (Edwards et al., 2014 [5]), providing a single-source, portable code base that runs efficiently on laptops, small homogeneous clusters, and the largest heterogeneous supercomputers currently available. To ensure scientific reliability, HARD includes a regression test suite that automatically reproduces canonical verification problems such as the Sod and LeBlanc shock tubes, and the Sedov blast wave, comparing numerical solutions against known analytical results. The project is distributed under an OSI-approved license, hosted on GitHub, and accompanied by reproducible build scripts and continuous integration workflows. This combination of performance portability, verification infrastructure, and community-focused development makes HARD a sustainable platform for advancing radiation hydrodynamics research across multiple domains.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2017

The effects of sampling location and turbulence on discharge estimates in short converging turbine intakes

Standards provide recommendations for best practices when installing current meters to measure fluid flow in closed conduits. A central guideline requires the velocity distribution to be regular and the flow steady. Because of the nature of the short converging intakes typical of low-head hydroturbines, these assumptions may be invalid if current meters are intended to be used to estimate discharge. Usual concerns are (1) the effects of the number of devices, (2) the sampling location and (3) the high turbulence caused by blockage from submersible traveling screens usually deployed for safe downstream fish passage. These three effects were examined in the present study by using 3D simulated flow fields in both steady-state and transient modes. In the process of describing an application at an existing hydroturbine intake at Ice Harbor Dam, the present work outlined the methods involved, which combined computational fluid dynamics, laboratory measurements in physical models of the hydroturbine, and current meter performance evaluations in experimental settings. The main conclusions in this specific application were that a steady-state flow field sufficed to determine the adequate number of meters and their location, and that both the transverse velocity and turbulence intensity had a small impact on estimate errors. However, while it may not be possible to extrapolate these findings to other field conditions and measuring devices, the study laid out a path to conduct similar assessments in other applications.

computational fluid dynamics (CFD)↗

DOE OSTI2023

Bentonite buffer under high temperature: laboratory experiments and coupled process modeling

Abstract. Bentonite buffer in a geological repository will be simultaneously heated from decaying radioactive waste and hydrated from the surrounding host rock, triggering complex and coupled THMC (thermal–hydrological–mechanical–chemical) processes. Understanding the THMC behavior of bentonite-based engineered barrier system (EBS) is key to the evaluation and prediction of its long-term performance. Studies on the THMC process have been focused on conditions under 100 ∘C, as most design concepts impose a thermal limit of 100 ∘C in bentonite. Recently, studies under high-temperature conditions have been conducted to evaluate the possibility of raising the thermal limit and expanding the data/knowledge base to increase the confidence level. In this abstract, we present a series of bench-scale laboratory experiments at high temperatures (up to 200 ∘C) and the corresponding modeling work. Two sets of column tests were conducted, and each set consisted of two test columns: a control column undergoing only hydration (non-heated) and an experiment column experiencing both heating and hydration (heated). During the experiment, frequent X-ray computed tomography (CT) images were collected to provide a 3D visualization of the density distribution and present the spatiotemporal evolution of (1) hydration/dehydration, (2) clay swelling/shrinkage, (3) displacement, and (4) mineral precipitation. The two sets of tests differ with respect to several experimental conditions, such as bentonite type, compacted density and water content, water chemistry, and hydration pressure, but the important difference is that the first set used bentonite powder with a dry density of 1.28 g cm−3, whereas the second set used granulated bentonite (mixture of pellets and powder) with a dry density of 1.45–1.5 g cm−3. In both sets of experiments, a comprehensive post-dismantling characterization of bentonite samples was carried out after the column tests had been running for 1.5 years. Comparing non-heated and heated columns, the temperature gradient led to lower degree of homogenization of bentonite after bentonite became fully saturated; comparing the first and second sets, granulated and powdered bentonite exhibited drastically different hydration behavior. A THM model with a 2D axisymmetric grid system was used to interpret the data from the first set of tests. The model considers the combined impact of saturation, fluid pressure, and porosity change due to swelling/compression on the spatiotemporal distribution of bulk density and movement of the thermocouple modules. Observations from the tests help us understand the early perturbation of bentonite buffer under high temperature, and data from these tests improve the calibration of key constitutive hydrological and mechanical models and, therefore, enhance the modeling capability with respect to calculating the long-term evolution of bentonite buffer.

Zheng, Liange↗

NASA NTRS1991

Wind tunnel blockage study of a generic three-dimensional sidewall compression scramjet inlet at Mach 10

A large scale model of a generic three-dimensional sidewall compression scramjet inlet has been designed based on the results of a computational parametric study for testing in the 31-inch Mach 10 Hypersonic Wind Tunnel at the NASA Langley Research Center. In order to increase the instrumentation density in interaction regions for a highly instrumented model, it is desirable to make the model as large as possible. When the cross-sectional area of a model becomes large relative to the inviscid core size of the tunnel, the effects of blockage must be considered. In order to assess these effects, a blockage model (an inexpensive, much less densely instrumented version of the configuration) was fabricated for preliminary testing. Since it was desired to determine both the effect of the model on the performance of the wind tunnel and also to determine if the inlet would start, the model possessed a total of 32 static pressure orifices distributed on the forebody plane and sidewalls; seventeen static pressure orifices on the tunnel wall and 3 pitot probes on the model monitored the tunnel performance. This paper presents the design considerations in the development of the wind tunnel model and the blockage aspects of the effects of contraction ratio, cowl location, Reynolds number, and angle of attack.

Holland, Scott D.↗

NASA NTRS1998

Monte Carlo Calculations of Polarized Microwave Radiation Emerging from Cloud Structures

The last decade has seen tremendous growth in cloud dynamical and microphysical models that are able to simulate storms and storm systems with very high spatial resolution, typically of the order of a few kilometers. The fairly realistic distributions of cloud and hydrometeor properties that these models generate has in turn led to a renewed interest in the three-dimensional microwave radiative transfer modeling needed to understand the effect of cloud and rainfall inhomogeneities upon microwave observations. Monte Carlo methods, and particularly backwards Monte Carlo methods have shown themselves to be very desirable due to the quick convergence of the solutions. Unfortunately, backwards Monte Carlo methods are not well suited to treat polarized radiation. This study reviews the existing Monte Carlo methods and presents a new polarized Monte Carlo radiative transfer code. The code is based on a forward scheme but uses aliasing techniques to keep the computational requirements equivalent to the backwards solution. Radiative transfer computations have been performed using a microphysical-dynamical cloud model and the results are presented together with the algorithm description.

Kummerow, Christian↗

NASA NTRS1997

Communication Studies of DMP and SMP Machines

Understanding the interplay between machines and problems is key to obtaining high performance on parallel machines. This paper investigates the interplay between programming paradigms and communication capabilities of parallel machines. In particular, we explicate the communication capabilities of the IBM SP-2 distributed-memory multiprocessor and the SGI PowerCHALLENGEarray symmetric multiprocessor. Two benchmark problems of bitonic sorting and Fast Fourier Transform are selected for experiments. Communication-efficient algorithms are developed to exploit the overlapping capabilities of the machines. Programs are written in Message-Passing Interface for portability and identical codes are used for both machines. Various data sizes and message sizes are used to test the machines' communication capabilities. Experimental results indicate that the communication performance of the multiprocessors are consistent with the size of messages. The SP-2 is sensitive to message size but yields a much higher communication overlapping because of the communication co-processor. The PowerCHALLENGEarray is not highly sensitive to message size and yields a low communication overlapping. Bitonic sorting yields lower performance compared to FFT due to a smaller computation-to-communication ratio.

Sohn, Andrew↗

DOE OSTI2021

Novel usage of deep learning and high-performance computing in long-baseline neutrino oscillation experiments

Mención Internacional en el título de doctorDeep-learning methods are playing a crucial role in numerous scientific and industrialapplications. Over the past two decades, these techniques have helped in the collection,reconstruction, and analysis of large data samples in particle physics experiments. Themain topic of this PhD research is the study of deep-learning techniques in long-baselineneutrino oscillation experiments. Neutrinos are mysterious light elementary particles,and their investigation is essential to shed light on some of the remaining open questionsin physics. The work presented here describes an algorithm based on a convolutionalneural network developed to provide highly accurate and efficient selections of electronneutrino and muon neutrino interactions in the Deep Underground Neutrino Experiment(DUNE). With this algorithm, the electron neutrino (antineutrino) selection efficiencypeaks at 90% (94%) and exceeds 85% (90%) for reconstructed neutrino energies between2-5 GeV. The selection efficiency for muon neutrino (antineutrino) interactions is foundto have a maximum of 96% (97%) and exceeds 90% (95%) efficiency for reconstructedneutrino energies above 2 GeV. When considering all electron neutrino and antineutrinointeractions as signal (both those appearing from oscillations and those intrinsic tothe beam), a selection purity of 90% is achieved. These event selections are criticalto maximise the sensitivity of the experiment to CP-violating effects, key to furtherunderstand the matter-antimatter asymmetry of the Universe.In high-energy physics experiments, deep learning has also been explored for producingfast simulations and physically-motivated manipulations of simulated images. Some ofthose simulations, such as the light production and detection, are very computationallyexpensive and require novel methods to produce the necessary samples while controllingthe varied underlying physics model parameters. To do so, we invented the model-assistedgenerative adversarial network (MAGAN), first validated on simple generic case studiesand then successfully applied to the DUNE photon-detector simulation.Moreover, we also developed graph neural networks for 3D-voxel classification ofambiguities and optical crosstalk for a different particle physics experiment, most preciselyfor the proposed SuperFGD. This novel 3D-granular plastic-scintillator neutrino detectorwill be used to upgrade the near detector of the T2K neutrino oscillation experiment, and our method reports efficiencies and purities of 94-96% per event in the classificationof particle track voxels.Due to the growth and complexity of deep neural networks, researchers have beeninvestigating techniques to train those networks in a more computationally-efficient way.Many efforts have been made by the community to optimise deep-learning models byparallelising or distributing their training computation across multiple devices. In thisthesis, we study an approach based on data locality for those neural networks that cannotbenefit from scaling their computation due to a significant bottleneck in the data I/O.The research also includes a detailed study on the performance of deep neural networkson hardware accelerator boards.Los métodos de aprendizaje profundo son cada vez más utilizados en numerosas aplicacionescientíficas e industriales hoy en día. Durante las dos últimas décadas, estastécnicas se han empleado en la recolección, reconstrucción y análisis de la gran cantidadde datos generados por experimentos de física de partículas. El tema principal de estatesis doctoral es el uso de estos modelos de aprendizaje profundo en experimentos defísica de neutrinos, en concreto en los experimentos de larga distancia DUNE y T2K. Losneutrinos, partículas fundamentales neutras, de las más ligeras del Universo, pueden serclave para explicar algunas de las cuestiones todavía sin resolver en física fundamental.Entre las diferentes contribuciones que esta tesis ha hecho a su estudio, cabe destacar eldesarrollo de un algoritmo basado en una red de neuronas convolucional para seleccionarcon gran eficiencia y precisión las interacciones de neutrinos electrónicos y muónicos enel Deep Underground Neutrino Experiment (DUNE). La eficiencia de selección obtenidapara neutrinos (antineutrinos) electrónicos alcanza un máximo del 90% (94%) y supera el85% (90%) para neutrinos con energías reconstruidas en el rango 2-5 GeV. La selección deneutrinos (antineutrinos) muónicos tiene una eficiencia máxima del 96% (97%) y excedeel 90% (95%) para neutrinos con energías reconstruidas de más de 2 GeV. Considerandocomo señal todas las interacciones de neutrinos y antineutrinos electrónicos (procedentestanto de oscilaciones como intrínsecos en el haz inicial), se logra una pureza en la seleccióndel 90%. Dichas selecciones de eventos son fundamentales para maximizar la sensibilidaddel experimento a los efectos de violació...

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗