Search NASA⌕ Search

SEARCH · Search NASA

Results for “hierarchical 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 649 records · Page 36

h5bench: A unified benchmark suite for evaluating HDF5 I/O performance on pre‐exascale platforms

Summary Parallel I/O is a critical technique for moving data between compute and storage subsystems of supercomputers. With massive amounts of data produced or consumed by compute nodes, high‐performant parallel I/O is essential. I/O benchmarks play an important role in this process; however, there is a scarcity of I/O benchmarks representative of current workloads on HPC systems. Toward creating representative I/O kernels from real‐world applications, we have created h5bench , a set of I/O kernels that exercise hierarchical data format version 5 (HDF5) I/O on parallel file systems in numerous dimensions. Our focus on HDF5 is due to the parallel I/O library's heavy usage in various scientific applications running on supercomputing systems. The various tests benchmarked in the h5bench suite include I/O operations (read and write), data locality (arrays of basic data types and arrays of structures), array dimensionality (one‐dimensional arrays, two‐dimensional meshes, three‐dimensional cubes), I/O modes (synchronous and asynchronous). In this paper, we present the observed performance of h5bench executed along several of these dimensions on existing supercomputers (Cori and Summit) and pre‐exascale platforms (Perlmutter, Theta, and Polaris). h5bench measurements can be used to identify performance bottlenecks and their root causes and evaluate I/O optimizations. As the I/O patterns of h5bench are diverse and capture the I/O behaviors of various HPC applications, this study will be helpful to the broader supercomputing and I/O community.

97 MATHEMATICS AND COMPUTING↗

A hierarchical analysis of the impact of methodological decisions on statistical downscaling of daily precipitation and air temperatures

Despite the widespread application of statistical downscaling tools, uncertainty remains regarding the role of model formulation in determining model skill for daily maximum and minimum temperature (Tmax and Tmin), and precipitation occurrence and intensity. Impacts of several key aspects of statistical transfer function form on model skill are evaluated using a framework resistant to model overspecification. We focus on: (a) model structure: simple (generalized linear models, GLMs) versus complex (artificial neural networks, ANNs) models. (b) Predictor selection: Fixed number of predictors chosen a priori versus stepwise selection of predictors and inclusion of grid point values versus predictors derived from application of principal components analysis (PCA) to spatial fields. We also examine the influence of domain size on model performance. For precipitation downscaling, we consider the role of the threshold used to characterize a wet day and apply three approaches (Poisson and Gamma distributions in GLM and ANN) to downscale wet-day precipitation amounts. While no downscaling formulation is optimal for all predictands and at 10 locations representing diverse U.S. climates, and due to the exclusion of variance inflation all of the downscaling formulations fail to reproduce the range of observed variability, models with larger suites of prospective predictors generally have higher skill. For temperature downscaling, ANNs generally outperform GLM, with greater improvements for T min than T max . Furthermore, use of PCA-derived predictors does not systematically improve model skill, but does improve skill for temperature extremes. Model skill for precipitation occurrence generally increases as the wet-day threshold increases and models using PCA-derived predictors tend to outperform those based on grid cell predictors. Each model for wet-day precipitation intensity overestimates annual total precipitation and underestimates the proportion derived from extreme precipitation events, but ANN-based models and those with larger predictor suites tend to have the smallest bias.

54 ENVIRONMENTAL SCIENCES↗

Probabilistic Groundwater Modeling of the RDX Plume at Los Alamos National Laboratory to Support Risk Assessment - 20359

A plume of the contaminant hexahydro-1,3,5-trinitro-1,3,5-triazine (RDX) with concentrations greater than the New Mexico tap water drinking standard (9.66 ppb) is present in the regional aquifer near the southwestern boundary of Los Alamos National Laboratory (LANL). A risk assessment is performed for exposure to regional aquifer groundwater, with long-term predictions of RDX concentrations provided by a calibrated, probabilistic, numerical fate and transport model. The structure of the model is hierarchical, with the RDX Regional Aquifer (RA) groundwater model acting as the primary tool for analysis of downgradient RDX concentrations. The RA model is deeply informed by the conceptual site model (CSM) and is calibrated using site RDX concentration data and hydraulic head measurements, along with other analyses. The model is calibrated using data through December 2019. Where data are scarce other lines of evidence are used to inform inputs, including the multiphase RDX Vadose Zone (VZ) model and Pipe and Disk analytical screening tool. Model inputs are described with informative prior distributions using a robust approach to development that incorporates all available lines of evidence for every parameter used as an input to the RA model. Calibration is performed using a classical nonlinear optimization routine, which is then used to initialize a Bayesian calibration. The Bayesian calibration constrains the uncertainty in the classical calibration, ultimately providing posterior distributions for all model parameters. The challenges of the calibration include high-dimensional parameter space, including spatially heterogeneous hydraulic conductivities, comparatively sparse data, and low RDX concentrations. Posterior parameter distributions developed in the calibration process are then used for stochastic predictive model runs into the future. The result of the forward runs is spatially and temporally explicit estimates of head and concentration with uncertainty at all points in space and time. The probabilistic modeling approach presented here includes innovative computational and statistical methods that leverage high-performance computing (HPC) resources. It makes use of prior modeling work performed at the chromium plume site in the central LANL area, with extensive updates. The risk assessment will ultimately be used to support decision-making at the site, using multiple appropriately weighted sources of information, as well as uncertainty, sensitivity, and value-of-information analyses. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Copper oxide assisted cysteine hierarchical structures for immunosensor application

The present work describes the promising electrochemical immunosensing strategy based on copper (II) assisted hierarchical cysteine structures (CuCys) varying from star to flower like morphology. The CuCys having average size of 10 μm have been synthesised using L-Cysteine as initial precursor in presence of copper oxide under environmentally friendly conditions in aqueous medium. To delineate the synthesis mechanism, detailed structural investigations have been carried out using characterization techniques such as X-ray diffraction, transmission electron microscopy, and Fourier transform infrared spectroscopy. The electrochemical behaviour of self-assembled CuCys on gold electrode shows surface controlled electrode reaction with an apparent electron transfer rate constant of 3.38 × 10{sup −4 }cm s{sup −1}. This innovative platform has been utilized to fabricate an immunosensor by covalently immobilizing monoclonal antibodies specific for Escherichia coli O157:H7 (E. coli). Under the optimal conditions, the fabricated immunosensor is found to be sensitive and specific for the detection of E. coli with a detection limit of 10 cfu/ml.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Geometric remapping of particle distributions in the Discrete Element Model for Sea Ice (DEMSI v0.0)

Abstract. A new sea ice dynamical core, the Discrete Element Model for Sea Ice (DEMSI), is under development for use in coupled Earth system models. DEMSI is based on the discrete element method, which models collections of ice floes as interacting Lagrangian particles. In basin-scale sea ice simulations the Lagrangian motion results in significant convergence and ridging, which requires periodic remapping of sea ice variables from a deformed particle configuration back to an undeformed initial distribution. At the resolution required for Earth system models we cannot resolve individual sea ice floes, so we adopt the sub-grid-scale thickness distribution used in continuum sea ice models. This choice leads to a series of hierarchical tracers depending on ice fractional area or concentration that must be remapped consistently. The circular discrete elements employed in DEMSI help improve the computational efficiency at the cost of increased complexity in the effective element area definitions for sea ice cover that are required for the accurate enforcement of conservation. An additional challenge is the accurate remapping of element values along the ice edge, the location of which varies due to the Lagrangian motion of the particles. In this paper we describe a particle-to-particle remapping approach based on well-established geometric remapping ideas that enforces conservation, bounds preservation, and compatibility between associated tracer quantities, while also robustly managing remapping at the ice edge. One element of the remapping algorithm is a novel optimization-based flux correction that enforces concentration bounds in the case of nonuniform motion. We demonstrate the accuracy and utility of the algorithm in a series of numerical test cases.

58 GEOSCIENCES↗

High-veracity functional imaging in scanning probe microscopy via Graph-Bootstrapping

The key objective of scanning probe microscopy (SPM) techniques is the optimal representation of the nanoscale surface structure and functionality inferred from the dynamics of the cantilever. This is particularly pertinent today, as the SPM community has seen a rapidly growing trend towards simultaneous capture of multiple imaging channels and complex modes of operation involving high-dimensional information-rich datasets, bringing forward the challenges of visualization and analysis, particularly for cases where the underlying dynamic model is poorly understood. To meet this challenge, we present a data-driven approach, Graph-Bootstrapping, based on low-dimensional manifold learning of the full SPM spectra and demonstrate its successes for high-veracity mechanical mapping on a mixed polymer thin film and resolving irregular hydration structure of calcite at atomic resolution. Using the proposed methodology, we can efficiently reveal and hierarchically represent salient material features with rich local details, further enabling denoising, classification, and high-resolution functional imaging.

36 MATERIALS SCIENCE↗

Sensitivity analysis in multipole-accelerated panel methods for potential flow

In the design of an airframe, the effect of changing the geometry on resulting computations is necessary for design optimization. The geometry is defined in terms of a series of design variables, including design variables to define the wing planform, tail, canard, pylon, and nacelle. Design optimization in this research is based on how these design variable affect the potential flow. The potential flow is computed as a function of the geometry and location of a series of panels describing the airframe, which are in turn a function of the design variables. Multipole accelerated panel methods improve the computational complexity of the problem and thus are an attractive approach. To utilize the methods in design optimization, it was necessary to define the appropriate sensitivity derivatives. The overhead incurred from finding the sensitivity derivatives in conjunction with the original computation should be small. This research developed the background for multipole-accelerated panel methods and the framework for finding sensitivity derivatives in the methods. Potential flow panel codes are commonly used for powered-lift aerodynamic predictions for three dimensional geometries. Given an airframe which has been discretized into a series of panels to define the airframe geometry, potential is computed as a function of the influence of all panels on all other panels. This is a computationally intensive problem for which efficient solutions are desired to improve the computational time and to allow greater resolution by use of more panels. One such solution is the use of hierarchical multipole methods which entail approximations of the effects of far-field terms. Hierarchical multipole methods have become prevalent in molecular dynamics and gravitational physics, and have been introduced into the fields of capacitance calculations, computational fluid dynamics, and electromagnetics. The methods utilize multipole expansions to describe the effect of bodies (i.e. particles, astrophysical bodies, panels, etc.) within a sphere on points distant from the sphere, where the influence diminishes as a function of distance. The expansions are exact with infinite series, however, for practical computations, the series are truncated and accuracy is selected based on the number of terms retained in the expansions. A hierarchical tree structure groups bodies together based on proximity to allow definition of multipole expansions for each group. The multipole expansions are then used to compute the effect of the bodies in a group on distant bodies.

Leathrum, James F., Jr.↗

Hierarchical electrode design of highly efficient and stable unitized regenerative fuel cells (URFCs) for long-term energy storage

The unitized regenerative fuel cell (URFC) is a promising electrochemical device for intermittent renewable energy storage in chemical bonds. However, widespread application has been hindered due to low round-trip efficiencies (RTEs) and disappointing durability, in particular at high rates. In this study, we break through that barrier by demonstrating highly efficient, flexible, and stable URFCs via hierarchical design of the multiscale catalyst-layer structures. A more porous and less tortuous Pt and Ir catalyst layer is realized using a doctor blade fabrication method that significantly improves URFC performance. We demonstrate RTEs of 56% and 53% under constant-electrode and constant-gas mode, respectively, while operating at 1000 mA cm –2 , and significantly, a RTE of 45% at 2000 mA cm –2 , achievements that were previously viewed as unfeasible under the onerous demands of URFC operation. At the same time we demonstrate URFCs under both constant-electrode and constant-gas mode operated continuously for over 500 h with negligible degradation. These results demonstrate the viability of applying URFCs for long-term energy storage at previously unattainable efficiencies and cast new light on electrode design and optimization of URFCs.

25 ENERGY STORAGE↗

Point-Mass Aircraft Trajectory Prediction Using a Hierarchical, Highly-Adaptable Software Design

A highly adaptable and extensible method for predicting four-dimensional trajectories of civil aircraft has been developed. This method, Behavior-Based Trajectory Prediction, is based on taxonomic concepts developed for the description and comparison of trajectory prediction software. A hierarchical approach to the "behavioral" layer of a point-mass model of aircraft flight, a clear separation between the "behavioral" and "mathematical" layers of the model, and an abstraction of the methods of integrating differential equations in the "mathematical" layer have been demonstrated to support aircraft models of different types (in particular, turbojet vs. turboprop aircraft) using performance models at different levels of detail and in different formats, and promise to be easily extensible to other aircraft types and sources of data. The resulting trajectories predict location, altitude, lateral and vertical speeds, and fuel consumption along the flight path of the subject aircraft accurately and quickly, accounting for local conditions of wind and outside air temperature. The Behavior-Based Trajectory Prediction concept was implemented in NASA's Traffic Aware Planner (TAP) flight-optimizing cockpit software application.

Karr, David A.↗

Small-Signal Angle Stability-Oriented False Data Injection Cyber-Attacks on Power Systems

The small-signal angle stability (SSAS) of a power system is determined by the property of operation points. The widely applied false data injection (FDI) cyber-attack, however, is able to stealthily mislead the optimal power flow (OPF) and thus compromise operation points, leading to damages to the SSAS margin. Here, to provide insights for cyber defenders, this paper proposes and investigates a stealthy SSAS-oriented FDI cyber-attack focusing on two attacking purposes, i.e., the SSAS margin and operation cost, with higher priority on the former one. First, this paper establishes a novel bi-level model with an implicit SSAS constraint based on a structure preserving model to compromise operation points. Then, for the SSAS interarea mode in a typical two-area system, this paper formulates closed-form expressions of how the SSAS margin and operation cost behave with respect to stealthy injections. By comparison, for the SSAS local mode in general power systems, this paper proposes a moving target cyber-attack-based hierarchical solution algorithm. Simulation results on a two-area system, a Kundur 11 bus system, and a modified IEEE 14 bus system demonstrate the significant damaging effects of the proposed SSAS-oriented FDI cyber-attack and the conflict between the two attacking purposes.

Benders decomposition↗

Formation and surface melting of nanoparticle superlattices in a solution

The wisdom in the saying of “There are no two snowflakes alike” lies in the importance of history or kinetic pathways in the phase transitions of solids. Likewise, “artificial solids,” namely superlattices consisting of functional nanoparticles, have lattice size, surface morphology, crystallinity, symmetry, and structural reconfiguration (for example, transition into a disordered state) highly dependent on the kinetic pathways as the nanoparticles interact with each other in solution [1]. Great progresses have been made in understanding the formation pathways of superlattices using liquid-phase transmission electron microscopy (TEM) [2-4]. For example, by tracking single nanoparticle’s trajectories, especially aided by U-net neural network-based machine learning, previous studies mapped the fundamental nanoparticle interactions at nanometer resolution [5]. Nonclassical, two-step nucleation pathway has also been elucidated in the system of nanoprisms, by optimizing protocols such as loading nanoparticle suspensions over the supersaturation threshold and minimizing particle‒substrate interaction [2]. Surface morphologies or exposed facets of superlattices have been shown to follow the principles of Wulff construction rule, where the facet-dependent surface energy can be measured based on the capillary wave theory [4]. However, the reverse process of crystallization of superlattices, the conversion from crystalline to disordered state, has been much less explored. On one hand, the melting of nanoparticle superlattices can provide a preferred pathway to induce structural reorganization or shuffling of building blocks for them to transform into different types of crystal structures. On the other hand, understanding nanoscale superlattice melting and comparing such behaviors with the prevailing surface melting theories developed for atomic/molecular solids can provide a potent way to engineer phase transitions of supra- and hierarchical structures constructed from nanoscale entities (e.g., DNA-coated nanoparticles, proteins), for their applications in reprogrammable and switchable materials with multifunctional properties [6, 7]. The experimental challenges to observe melting of superlattices are twofold. Practically it is difficult to load the initial superlattice form, in an intact manner, into the highly confined liquid-phase TEM chamber for in-situ observation. Here, the triggering of melting also needs meticulous manipulation of nanoparticle concentration, interparticle interaction, and solution environment.

Kim, Ahyoung↗

Additive friction stir deposition of SS316: Effect of process parameters on microstructure evolution

Solid state nature of additive friction stir deposition (AFSD) additive manufacturing process is very advantageous in terms of defect formation and microstructural refinement in the material. Current study presents the process optimization, microstructural evolution and kinetics of recrystallization for AFSD deposited low stacking fault energy material - SS316. As deposited microstructure shows equiaxed ultra fine grains with an average grain size of ~5.0 ± 0.5 μm. Shear deformation at high temperature during processing leads to the operation of restoration mechanisms. Observation of necklace type microstructure in the as deposited SS316 is attributed to discontinuous dynamic recrystallization during processing. Recrystallization kinetics of the AFSD SS316 is characterized using Johnson-Mehl-Avarami-Kolmogorov (JMAK) model. Deformation – thermal cycling during AFSD process resulted in inconsistent recrystallization kinetics. Variation in strain, strain rate and temperature during processing with processing parameters result in varying microstructure and tool wear. Further, high strength-ductility combination and sustained work hardening in as deposited SS316 appear to arise from transformation and twinning during deformation, leading to the formation of hierarchical twins and martensitic phase after deformation.

36 MATERIALS SCIENCE↗

Agentic traffic intelligence: Augmented human-in-the-loop scenario generation for microscopic traffic simulation

Traditional microscopic traffic simulation generation often relies on static datasets and manual design, limiting its ability to simulate complex conditions easily. This paper presents a novel framework, Agentic Traffic Intelligence, which combines human approval large language models (LLMs), the Real-Twin tool, and multi-agent systems to perform realistic microscopic traffic simulation scenario generation. The proposed framework incorporates human-in-the-loop (HIL) control, retrieval-augmented generation (RAG), and multi-agent control mechanisms. HIL mechanisms are used to guide multiple LLMs focused on attributes for microscopic simulation generation and to improve the interpretability and transparency of LLM execution for users. RAG enhances context extraction by dynamically integrating external knowledge sources for traffic scenario generation foundations. A multi-agent architecture with supervisory control coordinates the interaction of simulation components, including traffic simulators, control logic, and calibration tools. This enables the synthesis of simulation-ready scenarios that reflect dynamic demand profiles and behavior controls. Furthermore, the framework fuses multisource traffic data with unstructured context and supports iterative refinement through interactive user feedback. Validated through microscopic simulation using Simulation of Urban Mobility, the generated scenarios demonstrate high-fidelity network generation with inflow and turn movement and behavioral calibration, offering a robust and efficient tool for stress-testing and optimizing urban mobility systems.

Hierarchical multi-agent control↗

Ultrathin oxide layers for nanoscale integration of molecular light absorbers, catalysts, and complete artificial photosystems

Artificial photosynthesis is an attractive approach for the generation of renewable fuels because such systems will be suitable for deployment on highly abundant, non-arable land. Recently emerged methods of nanoscience to create conformal, ultrathin oxide layers enable the hierarchical integration of light absorbers, catalysts, and membranes into systems with far simpler synthetic approaches than available till now. This holds in particular for the coupling of molecular light absorbers and catalysts for sunlight to fuel conversion, providing photoelectrodes with greatly improved stability. Moreover, the use of ultrathin inert oxides as proton conducting, molecule impermeable membranes has opened up the integration of reduction and oxidation half reactions into complete photosynthetic systems on the shortest possible length scale-the nanometer scale. This capability affords minimization of energy-degrading resistance losses caused by ion transport over macroscale distances while separating the incompatible water oxidation and carbon dioxide reduction catalysis environments on the nanoscale. Understanding of charge transport between molecular components embedded in the oxide layers is critical for guiding synthetic design improvements of the light absorber-catalyst units to optimize performance and integrate them into complete artificial photosystems. Recent results and insights from transient optical, vibrational, and photoelectrochemical studies are presented, and future challenges and opportunities for engaging dynamic spectroscopies to accelerate the development of nanoscale integrated artificial photosystems are discussed.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Bridging length scales in hard materials with ultra-small angle X-ray scattering – a critical review

Owing to their exceptional properties, hard materials such as advanced ceramics, metals and composites have enormous economic and societal value, with applications across numerous industries. Understanding their microstructural characteristics is crucial for enhancing their performance, materials development and unleashing their potential for future innovative applications. However, their microstructures are unambiguously hierarchical and typically span several length scales, from sub-ångstrom to micrometres, posing demanding challenges for their characterization, especially for in situ characterization which is critical to understanding the kinetic processes controlling microstructure formation. This review provides a comprehensive description of the rapidly developing technique of ultra-small angle X-ray scattering (USAXS), a nondestructive method for probing the nano-to-micrometre scale features of hard materials. USAXS and its complementary techniques, when developed for and applied to hard materials, offer valuable insights into their porosity, grain size, phase composition and inhomogeneities. We discuss the fundamental principles, instrumentation, advantages, challenges and global status of USAXS for hard materials. Using selected examples, we demonstrate the potential of this technique for unveiling the microstructural characteristics of hard materials and its relevance to advanced materials development and manufacturing process optimization. We also provide our perspective on the opportunities and challenges for the continued development of USAXS, including multimodal characterization, coherent scattering, time-resolved studies, machine learning and autonomous experiments. Our goal is to stimulate further implementation and exploration of USAXS techniques and inspire their broader adoption across various domains of hard materials science, thereby driving the field toward discoveries and further developments.

36 MATERIALS SCIENCE↗

Benchmarks and Tests of a Multidimensional Cluster Dynamics Model of Helium Implantation in Tungsten

Here, we present a hierarchical multiscale modeling study of implanted helium (He) segregation near grain boundaries (GBs) of tungsten. We extend our spatially dependent cluster dynamics model to two spatial dimensions in order to take into account the biased drift of mobile He clusters toward the GBs observed in atomic-scale simulations. We are also able to reproduce the results from large-scale molecular dynamics simulations near and away from the GBs at low fluence with the extended cluster dynamics model. We suggest and verify that the sink (surface and GB) strengths are attenuated by the increasing concentration of He clusters at high fluence. This cluster dynamics model continues to set the stage for development of fully atomistically informed, coarse-grained models for computationally efficient predictions of He retention and surface morphological evolution, advancing progress toward the goal of efficient and optimal design of plasma-facing components.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

DNA-Enabled Chiral Gold Nanoparticle–Chromophore Hybrid Structure with Resonant Plasmon–Exciton Coupling Gives Unusual and Strong Circular Dichroism

Circular dichroism (CD) from hybrid complexes of plasmonic nanostructures and chiral molecules has recently attracted significant interest. Yet, the hierarchical chiral self-assembly of molecules on surfaces of metal nanostructures has remained challenging. As a result, a deep understanding of plasmon–exciton coupling between surface plasmons and chiral collective molecular excitations has not been achieved. In particular, the critical impact of resonant plasmon–exciton coupling within the hybrid is unclear. Here, we employed DNA-templated strategies to control the chiral self-assembly of achiral chromophores with rationally tuned exciton transitions on gold nanosphere (AuNP) or gold nanorod (AuNR) surfaces. Unlike many previous chiral plasmonic hybrids utilizing chiral biomolecules with CD signals in the UV range, we designed structures with the chiral excitonic resonances at visible wavelengths. The constructed hybrid complexes displayed strong chiroptical activity that depends on the spectral overlap between the chiral collective molecular excitations and the plasmon resonances. We find that when spectral overlap is optimized, the molecular CD signal originating from the chiral self-assemblies of chromophores was strongly enhanced (maximum enhancement of nearly an order of magnitude) and a plasmonic CD signal was induced. Surprisingly, the sign of the molecular CD was reversed despite different self-assembly mechanisms of the Au nanoparticle–chromophore hybrids. Our findings offer new insight into plasmonic CD enhancements and will inspire further studies on chiral light–matter interactions in strongly coupled plasmonic–excitonic systems.

77 NANOSCIENCE AND NANOTECHNOLOGY↗

A High-Throughput Solver for Marginalized Graph Kernels on GPU

Here, we present the design and optimization of a solver for efficient and high-throughput computation of the marginalized graph kernel on General Purpose GPUs. The graph kernel is computed using the conjugate gradient method to solve a generalized Laplacian of the tensor product between a pair of graphs. To cope with the large gap between the instruction throughput and the memory bandwidth of the GPUs, our solver forms the graph tensor product on-the-fly without storing it in memory. This is achieved by using threads in a warp cooperatively to stream the adjacency and edge label matrices of individual graphs by small square matrix blocks called tiles, which are then staged in registers and the shared memory for later reuse. Warps across a thread block can further share tiles via the shared memory to increase data reuse. We exploit the sparsity of the graphs hierarchically by storing only non-empty tiles using a coordinate format and nonzero elements within each tile using bitmaps. We propose a new partition-based reordering algorithm for aggregating nonzero elements of the graphs into fewer but denser tiles to further exploit sparsity. We carry out extensive theoretical analyses on the graph tensor product primitives for tiles of various density and evaluate their performance on synthetic and real-world datasets. Our solver delivers three to four orders of magnitude speedup over existing CPU-based solvers such as GraKeL and GraphKernels. The capability of the solver enables kernel-based learning tasks at unprecedented scales.

97 MATHEMATICS AND COMPUTING↗