Search NASA⌕ Search

SEARCH · Search NASA

Results for “image data parallel processing”

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 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

ParFlow Sand Tank: A tool for groundwater exploration

The ParFlow Sand Tank model is an open source application designed to allow users to interactively simulate and visualize groundwater movement through the subsurface. The app is designed for both research and education; teaching hydrogeology concepts and making it easy explore and run sophisticated groundwater simulations. Our goal is to support increased accessibility and usability of research grade hydrology tools for research and teaching. The Sand Tank application simulates groundwater and surface water fluxes as well as contaminant transport in real time using the integrated physical hydrology model ParFlow (Kollet & Maxwell, 2006; Maxwell & Miller, 2005; Osei-Kuffuor et al., 2014) and the particle tracking code EcoSlim (Maxwell et al., 2019). ParFlow is a numerical hydrology model that simulates spatially distributed groundwater and surface water flow. It is a well established research tool with more than 90 publications documenting its development use to advance our understanding of groundwater dynamics and groundwater surface water interactions from the hillslope to the continental scale e.g. (Condon et al., 2020; Condon & Maxwell, 2019; Maxwell & Condon, 2016). It is designed for efficient parallel computation and has been run on many platforms spanning from laptops to supercomputers. However, one of the challenges of ParFlow is that it requires significant training and hydrologic expertise to develop simulations. The Sand Tank application makes this model accessible to anyone for education and exploration. Our application uses ParFlow for its simulation backend and ParaView for the data loading and processing. The communication infrastructure relies on the ParaViewWeb framework. We use model templates deployed in Docker images to setup the Sand Tank framework. Users can build the application locally or interact with it through our web deployment. When interacting with a template users can interactively change model parameters like subsurface processes or pump/inject water into the subsurface and watch the system respond to their changes in real time as the simulation runs. Additionally, our template setup will allow more advanced users to build custom templates of increasing complexity for both research and educational purposes.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

Ghost imaging second harmonic generation microscopy

Second harmonic generation (SHG) microscopy is useful for visualizing interfaces and sub-structures within a wide range of materials due to the propensity for SHG to occur in non-centrosymmetric environments. However, since SHG is a nonlinear process generally necessitating small focal sizes for higher peak powers, a raster scanning approach is usually needed to build an SHG image over a significant sample size. While raster scanning is effective, there is a cost in terms of the time needed to acquire the image and, also, some materials cannot withstand the higher optical intensities within the small focal volume. As such, we describe a SHG microscopy approach based on ghost imaging (GI), which enables imaging data to be collected in parallel rather than sequentially as in raster scanning techniques. We experimentally demonstrate the approach and combine GI-SHG with compressive sensing to make further substantial gains in reducing the amount of sampling required for image reconstruction. Furthermore, GI-SHG is shown to have significant advantages for imaging in highly scattering environments, partly because GI is a background-free approach requiring spatial correlations between photons that travel two paths, with one path entirely devoid of sample interaction. This basic property of GI means that only the photons that travel unimpeded through the sample preserve the spatial correlations needed to reconstruct the image, while more scattered photons do not contribute to the overall GI signal. Finally, we compare the image quality and sampling properties of three different reconstruction algorithms used for compressive sensing.

47 OTHER INSTRUMENTATION↗
DOE OSTI2021

A physics-informed and hierarchically regularized data-driven model for predicting fluid flow through porous media

This paper presents a new deep learning data-driven model for predicting structure dependent pore-fluid velocity fields in rock. The model is based on a Convolutional Auto-Encoder (CAE) artificial neural network capable of learning from image data generated by direct numerical simulations of fluid flow through pore-structures, such as by Lattice Boltzmann or molecular dynamics methods. The main novelty of the model in comparison to previous CAE-based data-driven approaches consists of three parts. The first is a methodology for decomposing the full-domain of the porous media into sub-regions, or “sub-domains”, in order to reduce the overall size of the CAE, batch process the sub-domains in parallel, and enable the CAE to learn local and generalizable nonlinear mappings of pore-fluid velocities. The second consists of embedding the finite difference solutions of the incompressible Navier-Stokes and continuity equations into convolutional layers prior to the CAE in order to provide the CAE with knowledge of fluid dynamics physics (PhyFlow). The third main novelty is that the training of the CAE is regularized with a hierarchical loss function that encourages the learning of fluid flow patterns (in a way similar to ranked modes in principal component analysis), ranking from most to least important. This is shown to increase the stability in learning, reduce over-fitting, and promote interpretability of the CAE neural network layers (HierCAE). The comprehensive new data-driven model, which we call the PhyFlow-HierCAE model, is shown to exhibit improved accuracy and generalizability of flow field predictions over conventional CAE models, attributable to the embedded physical knowledge and the hierarchical regularization, as well as realize orders of magnitude speed-ups in computation times as a surrogate for the direct numerical simulations. Examples of training and forward predictions on unseen pore-structures are provided and evaluated for data from Lattice Boltzmann and molecular dynamics simulations of pore-fluid flow. The model is shown to be a fast and accurate emulator (or “surrogate”) for predicting effective permeability of unseen pore-structures based on learning from relatively small direct numerical simulation datasets.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2023

Correlative montage parallel array cryo-tomography for in situ structural cell biology

Abstract Imaging large fields of view while preserving high-resolution structural information remains a challenge in low-dose cryo-electron tomography. Here we present robust tools for montage parallel array cryo-tomography (MPACT) tailored for vitrified specimens. The combination of correlative cryo-fluorescence microscopy, focused-ion-beam milling, substrate micropatterning, and MPACT supports studies that contextually define the three-dimensional architecture of cells. To further extend the flexibility of MPACT, tilt series may be processed in their entirety or as individual tiles suitable for sub-tomogram averaging, enabling efficient data processing and analysis.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2021

Multichannel Analysis of Surface Waves Accelerated (MASWAccelerated): Software for efficient surface wave inversion using MPI and GPUs

Multichannel Analysis of Surface Waves (MASW) is a technique frequently used in geotechnical engineering and engineering geophysics to infer 1D layered models of seismic shear wave velocities in the top tens to hundreds of meters of the subsurface. We aim to accelerate MASW calculations by capitalizing on modern computer hardware available in the workstations of most engineers: multiple cores and graphics processing units (GPUs). We propose new parallel and GPU accelerated algorithms for computing 1D MASW inversion, and provide software implementations in C using Message Passing Interface (MPI) and CUDA. These algorithms take advantage of sparsity that arises in the problem, and the work balance between processes considers typical data trends. We compare our methods to an existing open source Matlab MASW tool. Our serial C implementation achieves a 2x speedup over the Matlab software, and we continue to see improvements by parallelizing the problem with MPI. Here we see nearly perfect strong and weak scaling for uniform data, and improve strong scaling for realistic data by repartitioning the problem to process mapping. By utilizing GPUs available on most modern workstations, we observe an additional 1.3x speedup over the serial C implementation on the first use of the method. We typically repeatedly evaluate theoretical dispersion curves as part of an optimization procedure, and on the GPU the kernel can be cached for faster reuse on later runs. We observe a 3.2x speedup on the cached GPU runs compared to the serial C runs. This work is the first open-source parallel or GPU-accelerated software tool for MASW imaging, and should enable geotechnical engineers to fully utilize all computer hardware at their disposal.

58 GEOSCIENCES↗
DOE OSTI2021

The Urban Observatory: A Multi-Modal Imaging Platform for the Study of Dynamics in Complex Urban Systems

We describe an “Urban Observatory” facility designed for the study of complex urban systems via persistent, synoptic, and granular imaging of dynamical processes in cities. An initial deployment of the facility has been demonstrated in New York City and consists of a suite of imaging systems—both broadband and hyperspectral—sensitive to wavelengths from the visible (∼400 nm) to the infrared (∼13 micron) operating at cadences of ∼0.01–30 Hz (characteristically ∼0.1 Hz). Much like an astronomical survey, the facility generates a large imaging catalog from which we have extracted observables (e.g., time-dependent brightnesses, spectra, temperatures, chemical species, etc.), collecting them in a parallel source catalog. We have demonstrated that, in addition to the urban science of cities as systems, these data are applicable to a myriad of domain-specific scientific inquiries related to urban functioning including energy consumption and end use, environmental impacts of cities, and patterns of life and public health. We show that an Urban Observatory facility of this type has the potential to improve both a city’s operations and the quality of life of its inhabitants.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

Integrated hydrogeophysical modelling and data assimilation for geoelectrical leak detection

Time-lapse electrical resistivity tomography (ERT) measurements provide indirect observations of hydrological processes in the Earth's shallow subsurface at high spatial and temporal resolution. ERT has been used in the past decades to detect leaks and monitor the evolution of associated contaminant plumes. Specifically, inverted resistivity images allow visualization of the dynamic changes in the structure of the plume. However, existing methods do not allow the direct estimation of leak parameters (e.g. leak rate, location, etc.) and their uncertainties. We propose an ensemble-based data assimilation framework that evaluates proposed hydrological models against observed time-lapse ERT measurements without directly inverting for the resistivities. Each proposed hydrological model is run through the parallel coupled hydro-geophysical simulation code PFLOTRAN-E4D to obtain simulated ERT measurements. The ensemble of model proposals is then updated using an iterative ensemble smoother. In this paper, we demonstrate the proposed framework on synthetic and field ERT data from controlled tracer injection experiments. Our results show that the approach allows joint identification of contaminant source location, initial release time, and solute loading from the cross-borehole time-lapse ERT data, alongside with an assessment of uncertainties in these estimates. We demonstrate a reduction in site-wide uncertainty by comparing the prior and posterior plume mass discharges at a selected image plane. This framework is particularly attractive to sites that have previously undergone extensive geological investigation (e.g., nuclear sites). It is well suited to complement ERT imaging and we discuss practical issues in its application to field problems.

58 GEOSCIENCES↗
DOE OSTI2023

Transforming microseismic clouds into near real-time visualization of the growing hydraulic fracture

SUMMARY Microseismic observations during unconventional reservoir stimulation are typically seen as a proxy for clusters of hydraulic fractures and the extent of the stimulated reservoir. Such straightforward interpretation is often misleading and fails to provide a physically reasonable image of the fracturing process. This paper demonstrates the application of a physics-based machine learning algorithm which enables a rapid and accurate fracture mapping from the microseismic data. Our training and validation data set relies on a history-matched geomechanical modelling workflow implemented in GEOS software for the Hydraulic Fracturing Test Site 1 (HFTS-1) project. For this study we augmented the simulated fracture growth through geostatistical modelling of induced seismicity, so that the synthetic microseismic catalogue matches the main statistical properties of the field observations. We formulated the problem of mapping the actual fracture in the clutter of events to parallel common video segmentation workflows: several past video frames (microseismic density snapshots) are passed through a deep convolutional network to classify whether a given voxel is associated with a fracture or intact rock. We found that for accurate fracture mapping, the network’s input and architecture must be augmented to incorporate the fluid injection parameters (pressure, rate, concentration of proppant, and location of the perforation within the cluster). The error rate for the network reached as little as 10 per cent of the fracture area, while a conventional microseismic interpretation approach yielded ∼300 per cent. Our approach also yields must faster predictions than conventional methods (minutes instead of weeks), and could enable engineers to make rapid decisions regarding engineering parameters (pumping rate, viscosity) in real time during stimulation.

58 GEOSCIENCES↗
DOE OSTI2022

FilDReaMS: II. Application to the analysis of the relative orientations between filaments and the magnetic field in four Herschel fields

Context. Both simulations and observations of the interstellar medium show that the study of the relative orientations between filamentary structures and the magnetic field can bring new insight into the role played by magnetic fields in the formation and evolution of filaments and in the process of star formation. Aims. We provide a first application of FilDReaMS, the new method presented in the companion paper to detect and analyze filaments in a given image. The method relies on a template that has the shape of a rectangular bar with variable width. Our goal is to investigate the relative orientations between the detected filaments and the magnetic field. Methods. We apply FilDReaMS to a small sample of four Herschel fields (G210, G300, G82, G202) characterized by different Galactic environments and different evolutionary stages. First, we look for the most prevalent bar widths, and we examine the networks formed by filaments of different bar widths as well as their hierarchical organization. Second, we compare the filament orientations to the magnetic field orientation inferred from Planck polarization data and, for the first time, we study the statistics of the relative orientation angle as functions of both spatial scale and H2 column density. Results. We find preferential relative orientations in the four Herschel fields: small filaments with low column densities tend to be slightly more parallel than perpendicular to the magnetic field; in contrast, large filaments, which all have higher column densities, are oriented nearly perpendicular (or, in the case of G202, more nearly parallel) to the magnetic field. In the two nearby fields (G210 and G300), we observe a transition from mostly parallel to mostly perpendicular relative orientations at an H 2 column density ≃ 1.1 × 10 21 cm -2 and 1.4 × 10 21 cm -2 , respectively, consistent with the results of previous studies. Conclusions. Our results confirm the existence of a coupling between magnetic fields at cloud scales and filaments at smaller scale. They also illustrate the potential of combining Herschel and Planck observations, and they call for further statistical analyses with our dedicated method.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Four-dimensional microstructurally small fatigue crack growth in cyclically loaded nickel superalloy specimen

Here, we nondestructively image a low-solvus high refractory (LSHR) nickel base specimen microstructure during cyclic loading to characterize crack propagation during the microstructurally small crack (MSC) growth regime. Micro-computed tomography (uCT) is used to characterize the material density and identify crack interfaces, while high-energy diffraction microscopy (HEDM) provides snapshots of the microstructural orientation field changes. From these measurements, data are computationally reconstructed and coregistered to describe the specimen’s configuration before load, before crack nucleation, and throughout the crack propagation/loading process. An MSC forms at a FIB-cut notch, which was previously placed to concentrate stress at the center of the specimen gauge, and propagates along a facet parallel to the local {1 1 1} crystallographic plane. Analysis of available crystallographic slip directions and measured point-to-point disorientation axes indicate probable single and coplanar slip activity along FCC octahedral planes within this region. Similar analyses of other crack propagation facets reveal generally slower growth along directions favoring zig-zag cross and codirectional cross slip, which are often disfavored at room temperature.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Quantitative spatiotemporal mapping of thermal runaway propagation rates in lithium-ion cells using cross-correlated Gabor filtering

Abuse testing of lithium-ion batteries is widely performed in order to develop new safety standards and strategies. However, testing methodologies are not standardised across the research community, especially with failure mechanisms being inherently difficult to reproduce. High-speed X-ray radiography is proven to be a valuable tool to capture events occurring during cell failure, but the observations made remain largely qualitative. We have therefore developed a robust image processing toolbox that can quantify, for the first time, the rate of propagation of battery failure mechanisms revealed by high-speed X-ray radiography. Using Gabor filter, the toolbox selectively tracks the electrode structure at the onset of failure. This facilitated the estimation of the displacement of electrodes undergoing abuse via nail penetration, and also the tracking of objects, such as the nail, as it propagates through a cell. Further, by cross-correlating the Gabor signals, we have produced practical, illustrative spatiotemporal maps of the failure events. From these, we can quantify the propagation rates of electrode displacement prior to the onset of thermal runaway. The highest recorded acceleration (≈514 mm s -2 ) was when a nail penetrated a cell radially (perpendicular to the electrodes) as opposed to axially (parallel to the electrodes). The initiation of thermal runaway was also resolved in combination with electrode displacement, which occurred at a lower acceleration (≈108 mm s -2 ). Our assistive toolbox can also be used to study other types of failure mechanisms, extracting otherwise unattainable kinetic data. Ultimately, this tool can be used to not only validate existing theoretical mechanical models, but also standardise battery failure testing procedures.

25 ENERGY STORAGE↗
DOE OSTI2025

Rapid characterization and failure analysis of 6276 rooftop-harvested photovoltaic connectors

Photovoltaic (PV) connectors, which link modules in series and connect PV strings in parallel, have increasingly been recognized as a primary contributor to PV system failures and a source of numerous fire incidents. However, publicly available data on the rates and types of connector failures are scarce, primarily due to the proprietary nature of the information and the need for comprehensive analysis. This study represents the first large-scale investigation of harvested PV connectors, drawing from a dataset of 6276 connectors from residential rooftop solar systems across the United States. The outcome of this work is twofold: 1) we have established a rapid characterization method for large populations of harvested connectors, incorporating visual inspection, resistance measurements, and X-ray imaging; and 2) the analysis made possible by our rapid-processing method has revealed, for a population of connector models provided by a single rooftop installer, failure statistics and insights for various connector makes and models, installation practices, operating currents, and internal component displacements. This research identifies common failure modes that could be considered in future connector designs standards, and operations and maintenance practices, to ultimately improve the reliability of this vital component of PV infrastructure.

MC4↗
DOE OSTI2025

The DECADE cosmic shear project I: A new weak lensing shape catalog of 107 million galaxies

We present the Dark Energy Camera All Data Everywhere (DECADE) weak lensing dataset: a catalog of 107 million galaxies observed by the Dark Energy Camera (DECam) in the northern Galactic cap. This catalog was assembled from public DECam data including survey and standard observing programs. These data were consistently processed with the Dark Energy Survey Data Management pipeline as part of the DECADE campaign and serve as the basis of the DECam Local Volume Exploration survey (DELVE) Early Data Release 3 (EDR3). We apply the Metacalibration measurement algorithm to generate and calibrate galaxy shapes. After cuts, the resulting cosmology-ready galaxy shape catalog covers a region of $5,\!412 \,\,{\rm deg}^2$ with an effective number density of $4.59\,\, {\rm arcmin}^{-2}$. The coadd images used to derive this data have a median limiting magnitude of $r = 23.6$, $i = 23.2$, and $z = 22.6$, estimated at ${\rm S/N} = 10$ in a 2 arcsecond aperture. We present a suite of detailed studies to characterize the catalog, measure any residual systematic biases, and verify that the catalog is suitable for cosmology analyses. In parallel, we build an image simulation pipeline to characterize the remaining multiplicative shear bias in this catalog, which we measure to be $m = (-2.454 \pm 0.124) \times10^{-2}$ for the full sample. Despite the significantly inhomogeneous nature of the data set, due to it being an amalgamation of various observing programs, we find the resulting catalog has sufficient quality to yield competitive cosmological constraints.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2024

Neural network methods for radiation detectors and imaging

Recent advances in image data proccesing through deep learning allow for new optimization and performance-enhancement schemes for radiation detectors and imaging hardware. This enables radiation experiments, which includes photon sciences in synchrotron and X-ray free electron lasers as a subclass, through data-endowed artificial intelligence. We give an overview of data generation at photon sources, deep learning-based methods for image processing tasks, and hardware solutions for deep learning acceleration. Most existing deep learning approaches are trained offline, typically using large amounts of computational resources. However, once trained, DNNs can achieve fast inference speeds and can be deployed to edge devices. A new trend is edge computing with less energy consumption (hundreds of watts or less) and real-time analysis potential. While popularly used for edge computing, electronic-based hardware accelerators ranging from general purpose processors such as central processing units (CPUs) to application-specific integrated circuits (ASICs) are constantly reaching performance limits in latency, energy consumption, and other physical constraints. These limits give rise to next-generation analog neuromorhpic hardware platforms, such as optical neural networks (ONNs), for high parallel, low latency, and low energy computing to boost deep learning acceleration (LA-UR-23-32395).

edge computing↗
DOE OSTI2022

Metaplastic and energy-efficient biocompatible graphene artificial synaptic transistors for enhanced accuracy neuromorphic computing

CMOS-based computing systems that employ the von Neumann architecture are relatively limited when it comes to parallel data storage and processing. In contrast, the human brain is a living computational signal processing unit that operates with extreme parallelism and energy efficiency. Although numerous neuromorphic electronic devices have emerged in the last decade, most of them are rigid or contain materials that are toxic to biological systems. In this work, we report on biocompatible bilayer graphene-based artificial synaptic transistors (BLAST) capable of mimicking synaptic behavior. The BLAST devices leverage a dry ion-selective membrane, enabling long-term potentiation, with ~50 aJ/µm 2 switching energy efficiency, at least an order of magnitude lower than previous reports on two-dimensional material-based artificial synapses. The devices show unique metaplasticity, a useful feature for generalizable deep neural networks, and we demonstrate that metaplastic BLASTs outperform ideal linear synapses in classic image classification tasks. With switching energy well below the 1 fJ energy estimated per biological synapse, the proposed devices are powerful candidates for bio-interfaced online learning, bridging the gap between artificial and biological neural networks.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Accelerating Binarized Neural Networks via Bit-Tensor-Cores in Turing GPUs

Despite foreseeing tremendous speedups over conventional deep neural networks, the performance advantage of binarized neural networks (BNNs) has merely been showcased on general-purpose processors such as CPUs and GPUs. In fact, due to being unable to leverage bit-level-parallelism with a word-based architecture, GPUs have been criticized for extremely low utilization (1%) when executing BNNs. Consequently, the latest tensorcores in NVIDIA Turing GPUs start to experimentally support bit computation. In this work, we look into this brand new bit computation capability and characterize its unique features. We show that the stride of memory access can significantly affect performance delivery and a data-format co-design is highly desired to support the tensorcores for achieving superior performance than existing software solutions without tensorcores. We realize the tensorcore-accelerated BNN design, particularly the major functions for fully-connect and convolution layers — bit matrix multiplication and bit convolution. Evaluations on two NVIDIA Turing GPUs show that, with ResNet-18, our BTC-BNN design can process ImageNet at a rate of 5.6K images per second, 77% faster than state-of-the-art. Our BNN approach is released on https://github.com/pnnl/TCBNN.

Li, Ang↗
DOE OSTI2024

Pretraining Billion-Scale Geospatial Foundational Models on Frontier

As AI workloads increase in scope, generalization capability becomes challenging for small task-specific models and their demand for large amounts of labeled training samples increases. On the contrary, Foundation Models (FMs) are trained with internet-scale unlabeled data via self-supervised learning and have been shown to adapt to various tasks with minimal fine-tuning. Although large FMs have demonstrated significant impact in natural language processing and computer vision, efforts toward FMs for geospatial applications have been restricted to smaller size models, as pretraining larger models requires very large computing resources equipped with state-of-the-art hardware accelerators. Current satellite constellations collect 100+TBs of data a day, resulting in images that are billions of pixels and multimodal in nature. Such geospatial data poses unique challenges opening up new opportunities to develop FMs. We investigate billion scale FMs and HPC training profiles for geospatial applications by pretraining on publicly available data. We studied from end-to-end the performance and impact in the solution by scaling the model size. Our larger 3B parameter size model achieves up to 30% improvement in top1 scene classification accuracy when comparing a 100M parameter model. Moreover, we detail performance experiments on the Frontier supercomputer, America's first exascale system, where we study different model and data parallel approaches using PyTorch's Fully Sharded Data Parallel library. Specifically, we study variants of the Vision Transformer architecture (ViT), conducting performance analysis for ViT models with size up to 15B parameters. By discussing throughput and performance bottlenecks under different parallelism configurations, we offer insights on how to leverage such leadership-class HPC resources when developing large models for geospatial imagery applications.

Tsaris, Aristeidis (aris)↗
DOE OSTI2020

AWB-GCN: A Graph Convolutional Network Accelerator with Runtime Workload Rebalancing

The recent development of deep learning has been mostly focusing on Euclidean data, such as images, videos, audios, etc. However, most real-world information and relation are often expressed as graphs. To efficiently learn from graph data, graph convolutional networks (GCNs) emerge as a promising approach, showing advantages in several practical applications such as social network analysis, knowledge discovery, 3D modeling, motion capturing, etc. Real-world graphs are usually extremely large and imbalanced, posting significant performance demand and design challenges on the hardware dedicated for GCN inference. In this paper, we propose an architecture design called UW-GCN to accelerate graph convolutional network inference. To tackle the major performance bottleneck from workload imbalance, we propose dynamic neighborhood stealing and remote chunk shuffling techniques, relying on hardware flexibility to achieve hardware auto-tuning under negligible area or delay overhead. Specifically, UW-GCN is able to smartly profile the sparse graph pattern while continuously adjusting the workload distribution via routing reconfiguration among parallel processing elements (PEs). The ideal configuration is then reused in the remaining iterations. To the best of our knowledge, this is the first accelerator design particularly for GCN and the first work relying on hardware auto-tuning, which is normally based on software, to achieve near-optimal workload balance in processing sparse structures.

Geng, Tong↗