Search NASA⌕ Search

SEARCH · Search NASA

Results for “3D neural networks”

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 55 records · Page 3
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Fast 2D Bicephalous Convolutional Autoencoder for Compressing 3D Time Projection Chamber Data

High-energy large-scale particle colliders produce data at high speed in the order of 1 terabytes per second in nuclear physics and petabytes per second in high energy physics. Developing real-time data compression algorithms to reduce such data at high throughput to fit permanent storage has drawn increasing attention. Specifically, at the newly constructed sPHENIX experiment at the Relativistic Heavy Ion Collider (RHIC), a time projection chamber is used as the main tracking detector, which records particle trajectories in a volume of three-dimensional (3D) cylinder. The resulting data are usually very sparse with occupancy around 10.8%. Such sparsity presents a challenge to conventional learning-free lossy compression algorithms, such as SZ, ZFP, and MGARD. The 3D convolutional neural network (CNN)-based approach, Bicephalous Convolutional Autoencoder (BCAE), outperforms traditional methods both in compression rate and reconstruction accuracy. BCAE can also utilize the computation power of graphical processing units suitable for deployment in a modern heterogeneous highperformance computing environment. This work introduces two BCAE variants: BCAE++ and BCAE-2D. BCAE++ achieves a 15% better compression ratio and a 77% better reconstruction accuracy measured in mean absolute error compared with BCAE. BCAE-2D treats the radial direction as the channel dimension of an image, resulting in a 3× speedup in compression throughput. In addition, we demonstrate an unbalanced autoencoder with a larger decoder can improve reconstruction accuracy without significantly sacrificing throughput. Lastly, we observe both the BCAE++ and BCAE-2D can benefit more from using half-precision mode in throughput (76 - 79% increase) without loss in reconstruction accuracy. The source code and links to data and pretrained models can be found at https://github.com/BNL-DAQ-LDRD/NeuralCompression_v2

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2022

Tomo2Mesh: Fast Reconstruction and Visualization of Tomography Data in Mesh Format

Tomo2Mesh is an open-source project targeted towards real-time reconstruction, segmentation, and visualization of computed tomography (CT) data in mesh format. The CT reconstruction scheme is based on filtered back-projection of voxel subsets. The segmentation scheme uses a 3D convolutional neural network. To allow for fast, real-time reconstruction, voxel subsets are first identified by coarse reconstruction. The detail in specific regions of interest is improved through full reconstruction of voxel subsets in that region. Data structures are implemented to store and process voxel subsets. Finally, voxel data is labeled using connected components to detect disconnected regions such as voids whose morphological attributes (e.g., Feret diameter, principal axis orientation, local number density, size, etc.) can be measured also in real-time. Finally, a fast marching cubes implementation processes labeled voxel data into a triangular face mesh (vertices and faces) in .ply format for visualization in Paraview or other mesh visualization tools. The code provides a simple programming interface for detecting, classifying, and visualizing regions of interest based on morphology. For example, detected voids can be classified as round pores or extended cracks. Highly porous neighborhoods can be identified based on local number density. The mesh texture (or color) is assigned based these morphological attributes to allow smart visualization scenarios in real-time (e.g., show only long cracks). At the time of first release (July 2022), extraction of face mesh for visualization for raw CT data from a 2 megapixel camera would take between 1-5 minutes for most scenarios.

TEKAWADE, ANIKET↗
DOE OSTI2020

A Hybrid Deep Learning Approach to Cosmological Constraints from Galaxy Redshift Surveys

We present a deep machine learning (ML)–based technique for accurately determining σ g and Ω m from mock 3D galaxy surveys. The mock surveys are built from the AbacusCosmos suite of N -body simulations, which comprises 40 cosmological volume simulations spanning a range of cosmological parameter values, and we account for uncertainties in galaxy formation scenarios through the use of generalized halo occupation distributions (HODs). We explore a trio of ML models: a 3D convolutional neural network (CNN), a power spectrum–based fully connected network, and a hybrid approach that merges the two to combine physically motivated summary statistics with flexible CNNs. We describe best practices for training a deep model on a suite of matched-phase simulations, and we test our model on a completely independent sample that uses previously unseen initial conditions, cosmological parameters, and HOD parameters. Despite the fact that the mock observations are quite small (~0.07 h -3 Gpc 3 ) and the training data span a large parameter space (six cosmological and six HOD parameters), the CNN and hybrid CNN can constrain estimates of σ g and Ω m to ~3% and ~4%, respectively.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Combining protein sequences and structures with transformers and equivariant graph neural networks to predict protein function

Abstract Motivation Millions of protein sequences have been generated by numerous genome and transcriptome sequencing projects. However, experimentally determining the function of the proteins is still a time consuming, low-throughput, and expensive process, leading to a large protein sequence-function gap. Therefore, it is important to develop computational methods to accurately predict protein function to fill the gap. Even though many methods have been developed to use protein sequences as input to predict function, much fewer methods leverage protein structures in protein function prediction because there was lack of accurate protein structures for most proteins until recently. Results We developed TransFun—a method using a transformer-based protein language model and 3D-equivariant graph neural networks to distill information from both protein sequences and structures to predict protein function. It extracts feature embeddings from protein sequences using a pre-trained protein language model (ESM) via transfer learning and combines them with 3D structures of proteins predicted by AlphaFold2 through equivariant graph neural networks. Benchmarked on the CAFA3 test dataset and a new test dataset, TransFun outperforms several state-of-the-art methods, indicating that the language model and 3D-equivariant graph neural networks are effective methods to leverage protein sequences and structures to improve protein function prediction. Combining TransFun predictions and sequence similarity-based predictions can further increase prediction accuracy. Availability and implementation The source code of TransFun is available at https://github.com/jianlin-cheng/TransFun.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2020

We Know Where We Don't Know: 3D Bayesian CNNs for Credible Geometric Uncertainty

Deep learning has been successfully applied to the segmentation of 3D Computed Tomography (CT) scans. Establishing the credibility of these segmentations requires uncertainty quantification (UQ) to identify untrustworthy predictions. Recent UQ architectures include Monte Carlo dropout networks (MCDNs), which approximate deep Gaussian processes, and Bayesian neural networks (BNNs), which learn the distribution of the weight space. BNNs are advantageous over MCDNs for UQ but are thought to be computationally infeasible in high dimension, and neither architecture has produced interpretable geometric uncertainty maps. We propose a novel 3D Bayesian convolutional neural network (BCNN), the first deep learning method which generates statistically credible geometric uncertainty maps and scales for application to 3D data. We present experimental results on CT scans of graphite electrodes and laser-welded metals and show that our BCNN outperforms an MCDN in recent uncertainty metrics. The geometric uncertainty maps generated by our BCNN capture distributions of sigmoid values that are interpretable as confidence intervals, critical for applications that rely on deep learning for high-consequence decisions.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Computationally Efficient Multiscale Neural Networks Applied to Fluid Flow in Complex 3D Porous Media

Abstract The permeability of complex porous materials is of interest to many engineering disciplines. This quantity can be obtained via direct flow simulation, which provides the most accurate results, but is very computationally expensive. In particular, the simulation convergence time scales poorly as the simulation domains become less porous or more heterogeneous. Semi-analytical models that rely on averaged structural properties (i.e., porosity and tortuosity) have been proposed, but these features only partly summarize the domain, resulting in limited applicability. On the other hand, data-driven machine learning approaches have shown great promise for building more general models by virtue of accounting for the spatial arrangement of the domains’ solid boundaries. However, prior approaches building on the convolutional neural network (ConvNet) literature concerning 2D image recognition problems do not scale well to the large 3D domains required to obtain a representative elementary volume (REV). As such, most prior work focused on homogeneous samples, where a small REV entails that the global nature of fluid flow could be mostly neglected, and accordingly, the memory bottleneck of addressing 3D domains with ConvNets was side-stepped. Therefore, important geometries such as fractures and vuggy domains could not be modeled properly. In this work, we address this limitation with a general multiscale deep learning model that is able to learn from porous media simulation data. By using a coupled set of neural networks that view the domain on different scales, we enable the evaluation of large ( $$>512^3$$ > 512 3 ) images in approximately one second on a single graphics processing unit. This model architecture opens up the possibility of modeling domain sizes that would not be feasible using traditional direct simulation tools on a desktop computer. We validate our method with a laminar fluid flow case using vuggy samples and fractures. As a result of viewing the entire domain at once, our model is able to perform accurate prediction on domains exhibiting a large degree of heterogeneity. We expect the methodology to be applicable to many other transport problems where complex geometries play a central role.

36 MATERIALS SCIENCE↗
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 ↗
DOE OSTI2020

Distributed fiber sensing systems for 3D combustion temperature field monitoring in coal-fired boilers using optically generated acoustic waves (Final Report)

In this project, we have developed and tested three kinds of fiber optic sensing systems for real time monitoring of temperature variations within an industrial scale boiler furnace. The fiber optic sensing systems target spatial and temporal distributions of high temperature profiles in a boiler furnace in fossil power plants. The reconstructed temperature profile will provide critical input for the control mechanisms to optimize the combustion process. This temperature profile will address the essential problem for fossil power plants in achieving higher efficiency and fewer pollutant emissions. Acoustic pyrometer systems have been used to reconstruct temperature field of power plant boilers based on measuring TOF (times-of-flight) of sound waves along some straight paths in a 2D cross-section of the boiler. In this project, optically generated acoustic signals from a fiber optic sensing system have replaced the acoustic signals generated from an electrical transducer. A 3D reconstruction algorithm replaced the previous 2D model. In this project, three kinds of fiber optic sensing systems have been developed and tested. They are fiber optic sensing system I, fiber optic sensing system II (Distributed Sensing System I) and fiber optic sensing system III (Distributed Sensing System II). For fiber optic sensing system I, the fiber optic ultrasound generator acts as a signal generator. A microphone, hydrophone or other electronic devices serve as a signal receiver. In this system, there are one generator and one receiver. Distance test, water temperature test, air temperature test, air temperature reconstruction, and GE ISBF pilot test were performed by Fiber optic sensing system I. The fiber optic sensing system I successfully detected temperature in all these tests. We got 2D temperature reconstruction results by using the fiber optic sensing system I and it matched the reference data. The fiber optic sensing system I successfully survived in GE ISBF boiler environment (480 °F). For fiber optic sensing system II (Distributed Sensing System I), it is an all optical ultrasound system. The fiber optic ultrasound generator acts as a signal generator. Fiber Bragg Grating (FBG) and Fabry-Perot (FP) sensor act as a signal receiver. In this system, there is one generator and one receiver. Aluminum plate temperature test, furnace high temperature test, and GE ISBF pilot test were performed by the fiber optic sensing system II. Fiber optic sensing system II successfully detected the temperature in all these tests. The fiber optic sensing system II successfully survived at up to 700 °C furnace environment and 320 °C GE ISBF boiler environment. For fiber optic sensing system III (Distributed Sensing System II), it is also an all optical ultrasound system. The fiber optic ultrasound generator acts as a signal generator. Multiple FP fiber sensors act as signal receivers. In this system, there is one generator and three receivers. Three GE ISBF pilot tests were performed by fiber optic sensing system III. The fiber optic sensing system III survived in the cold flow tests in GE’s ISBF pilot test facility. However, we didn’t get high temperature data by using this system since the nanosecond laser issues. During the period of the project, test trials, data simulation and algorithm optimization was performed successfully. For real time temperature field construction, the sampling rate must be fast enough to capture the field variations. The technology of Code-division multiple access (CDMA) is well studied which could allow parallel multiplexing, even if signals overlap in time or frequencies. Moreover, it has been known that extending the length of signal significantly improves SNR. For acoustic signals, these multiplexing techniques have also been widely used, mainly for sonar and acoustic communications. The CDMA modulation technique has been proposed and studied to guarantee high network throughput, low channel access delay and low energy consumption. We have studied the temperature field reconstruction using Gaussian Radial Basis Functions (GRBF)-based approximation approach. Reconstruction of 3D temperature field using Neural Networks with measured TOF and known propagation paths is feasible. 2D and 3D temperature field reconstruction simulation results are achieved. The milestone status is shown in Table 1. We finished milestone 1-8 and milestone 10. For milestone 9, we did three pilot tests by using the fiber optic sensing system III (Distributed Sensing System II) at GE Power. However, due to the failure of the ns laser, we did not get the temperature results. We conducted some additional tasks that were not originally proposed: 1) We fabricated a fiber optic sensing system I and did a pilot test based on this system. 2) In the proposal, we proposed two pilot tests at GE Power. In reality, we finished at least seven pilot tests at GE Power. GE Power has made a lot of efforts for supporting the pilot tests. 3) We got a simulation results based on CDMA. In summary, most of the tasks have been accomplished. The outcome of this project removed a few barriers that hinder the achievement of the final product of the distributed sensing systems. With the successful accomplishment of this project, a prototype of the fiber optic sensing system can be fabricated to attract more interests from companies and other funding agencies.

47 OTHER INSTRUMENTATION↗
DOE OSTI2021

The Application of Convolutional Neural Networks (CNNs) to Recognize Defects in 3D-Printed Parts

Cracks and pores are two common defects in metallic additive manufacturing (AM) parts. In this paper, deep learning-based image analysis is performed for defect (cracks and pores) classification/detection based on SEM images of metallic AM parts. Three different levels of complexities, namely, defect classification, defect detection and defect image segmentation, are successfully achieved using a simple CNN model, the YOLOv4 model and the Detectron2 object detection library, respectively. The tuned CNN model can classify any single defect as either a crack or pore at almost 100% accuracy. The other two models can identify more than 90% of the cracks and pores in the testing images. In addition to the application of static image analysis, defect detection is also successfully applied on a video which mimics the AM process control images. The trained Detectron2 model can identify almost all the pores and cracks that exist in the original video. This study lays a foundation for future in situ process monitoring of the 3D printing process.

36 MATERIALS SCIENCE↗
DOE OSTI2021

e3nn tutorial v0.2

Repository and website for tutorials on 3d Euclidean equivariant neural networks.

Smidt, Tess↗
DOE OSTI2023

The Feasibility of Incorporating a 3D Velocity Model Into Earthquake Location Around Salt Lake City, UT Using a Physics Informed Neural Network

Earthquake location algorithms typically require travel time calculation. Doing this calculation in 3D, despite advances in algorithm efficiency and computational power, can still be prohibitively expensive in terms of resources and storage. Implementation of high-resolution 3D models in routine earthquake location would be a significant step forward in most of the world. Machine learning algorithms have potential to act as substitutes for travel time calculation algorithms or stored travel time tables. We investigate EikoNet - a physics informed neural network machine learning model that estimates travel times very quickly and comes with negligible memory-overhead. Specifically, we apply EikoNet to the Wasatch Fault Community Velocity Model (WFCVM), a highly detailed and complex 3D velocity model of the Salt Lake City, UT region. While routine locations in the area and studies of the 2020 Magna, UT earthquake sequence used a 1D velocity model, a 3D model may help better our understanding the structure of the major fault in the region. Our primary goal was to test the speed, memory requirements, and accuracy of EikoNet compared to a reference eikonal solver. We find that while the EikoNet is exceedingly fast and requires little memory overhead, achieving acceptable accuracy in estimated travel times is difficult and requires extensive computational resources.

58 GEOSCIENCES↗
DOE OSTI2021

Rapid 3D nanoscale coherent imaging via physics-aware deep learning

Phase retrieval, the problem of recovering lost phase information from measured intensity alone, is an inverse problem that is widely faced in various imaging modalities ranging from astronomy to nanoscale imaging. The current process of phase recovery is iterative in nature. As a result, the image formation is time consuming and computationally expensive, precluding real-time imaging. Here, we use 3D nanoscale X-ray imaging as a representative example to develop a deep learning model to address this phase retrieval problem. We introduce 3D-CDI-NN, a deep convolutional neural network and differential programing framework trained to predict 3D structure and strain, solely from input 3D X-ray coherent scattering data. Our networks are designed to be “physics-aware” in multiple aspects; in that the physics of the X-ray scattering process is explicitly enforced in the training of the network, and the training data are drawn from atomistic simulations that are representative of the physics of the material. We further refine the neural network prediction through a physics-based optimization procedure to enable maximum accuracy at lowest computational cost. 3D-CDI-NN can invert a 3D coherent diffraction pattern to real-space structure and strain hundreds of times faster than traditional iterative phase retrieval methods. Our integrated machine learning and differential programing solution to the phase retrieval problem is broadly applicable across inverse problems in other application areas.

77 NANOSCIENCE AND NANOTECHNOLOGY↗
DOE OSTI2023

Hierarchical reconstruction of 3D well-connected porous media from 2D exemplars using statistics-informed neural network

The relationships between porous microstructures and transport properties are of fundamental importance in various scientific and engineering applications. Due to the intricacy, stochasticity and heterogeneity of porous media, reliable characterization and modeling of transport properties often require a complete dataset of internal microstructure samples. However, it is often an unbearable cost to acquire sufficient 3D digital microstructures by purely using microscopic imaging systems. Herein this paper presents a machine learning-based technique to hierarchically reconstruct 3D well-connected porous microstructures from one isotropic or several anisotropic low-cost 2D exemplar(s). To compactly characterize the large-scale microstructural features, a Gaussian image pyramid is built for each 2D exemplar. Local morphology patterns are collected from the Gaussian image pyramids, and then they serve as the training data to embed the 2D morphological statistics into feed-forward neural networks at multiple length levels. By using a specially-developed morphology integration scheme, the 3D morphological statistics at different levels can be inferred from the statistics-informed neural networks. Gibbs sampling is adopted to hierarchically reconstruct 3D microstructures by using multi-level 3D morphological statistics, where the large-scale, regional and local morphological patterns are statistically generated and successively added to the same 3D random field. The proposed method is tested on a series of porous media with distinct morphologies, and the statistical equivalence between the reconstructed and the real microstructures is systematically evaluated by comparing morphological descriptors and transport properties. The results demonstrate that the proposed 2D-to-3D microstructure reconstruction method is a universal and efficient approach to generating morphologically and physically realistic samples of porous media.

42 ENGINEERING↗
DOE OSTI2023

Machine learning for detection of 3D features using sparse x-ray tomographic reconstruction

In many inertial confinement fusion (ICF) experiments, the neutron yield and other parameters cannot be completely accounted for with one and two dimensional models. This discrepancy suggests that there are three dimensional effects that may be significant. Sources of these effects include defects in the shells and defects in shell interfaces, the fill tube of the capsule, and the joint feature in double shell targets. Due to their ability to penetrate materials, x rays are used to capture the internal structure of objects. Methods such as computational tomography use x-ray radiographs from hundreds of projections, in order to reconstruct a three dimensional model of the object. In experimental environments, such as the National Ignition Facility and Omega-60, the availability of these views is scarce, and in many cases only consists of a single line of sight. Mathematical reconstruction of a 3D object from sparse views is an ill-posed inverse problem. These types of problems are typically solved by utilizing prior information. Neural networks have been used for the task of 3D reconstruction as they are capable of encoding and leveraging this prior information. We utilize half a dozen, different convolutional neural networks to produce different 3D representations of ICF implosions from the experimental data. Deep supervision is utilized to train a neural network to produce high-resolution reconstructions. These representations are used to track 3D features of the capsules, such as the ablator, inner shell, and the joint between shell hemispheres. Machine learning, supplemented by different priors, is a promising method for 3D reconstructions in ICF and x-ray radiography, in general.

Wolfe, Bradley T. (ORCID:0000000268301614)↗
DOE OSTI2020

3D-ReG: A 3D ReRAM-based Heterogeneous Architecture for Training Deep Neural Networks

Deep neural network (DNN) models are being expanded to a broader range of applications. The computational capability of traditional hardware platforms cannot accommodate the growth of model complexity. Among recent technologies to accelerate DNN, resistive memory (ReRAM)-based processing-in-memory (PIM) emerged as a promising solution for DNN inference due to its high efficiency for matrix-based computation. We face two major technical challenges in extending the use of ReRAM-based accelerators for training: (1) full-precision data is essential in back-propagation; (2) the need to support both feed-forward and back-propagation aggravates the data-movement burden. We propose a heterogeneous architecture named as 3D-ReG, which leverages full-precision GPU to ensure training accuracy and low-overhead 3D integration to provide low-cost data movements. Moreover, we introduce conservative and aggressive task-mapping schemes, which partition the computation phases in different ways to balance execution efficiency and training accuracy. We evaluate 3D-ReG implemented with two 3D integration technologies, through-silicon vias (TSVs) and monolithic inter-tier vias (MIVs), and compare them with GPU-only and PIM-only counterparts. Various GPU-only platforms using two main-memory technologies (DRAM, ReRAM) and three interconnect technologies (2D, TSV, MIV) are evaluated as well. Experimental results show that 3D-ReG can achieve on average 5.64× training speedup and 3.56× higher energy efficiency compared with the GPU with DRAM as main memory, at the cost of 0.05%–3.39% accuracy drop. We define a new metric, gain-loss ratio (GLR), which quantitatively evaluates the capability of a DNN training hardware in terms of the model accuracy and hardware efficiency. The results of our comparison show that the aggressive task-mapping scheme on MIV-based 3D-ReG outperforms the other methods.

Computer Science↗
DOE OSTI2023

Deep learning enables nanoscale X-ray 3D imaging with limited data

Deep neural network can greatly improve tomography reconstruction with limited data. A recent effort of combining ptycho-tomography model with the 3D U-net demonstrated a significant reduction in both the number of projections and computation time, and showed its potential for integrated circuit imaging that requires high-resolution and fast measurement speed.

47 OTHER INSTRUMENTATION↗