Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Neural Networks (Computer)”

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 1,531 records · Page 85

Train Like a (Var)Pro: Efficient Training of Neural Networks with Variable Projection

Deep neural networks (DNNs) have achieved state-of-the-art performance across a variety of traditional machine learning tasks, e.g., speech recognition, image classification, and segmentation. The ability of DNNs to efficiently approximate high-dimensional functions has also motivated their use in scientific applications, e.g., to solve partial differential equations and to generate surrogate models. In this paper, we consider the supervised training of DNNs, which arises in many of the above applications. We focus on the central problem of optimizing the weights of the given DNN such that it accurately approximates the relation between observed input and target data. Devising effective solvers for this optimization problem is notoriously challenging due to the large number of weights, nonconvexity, data sparsity, and nontrivial choice of hyperparameters. To solve the optimization problem more efficiently, we propose the use of variable projection (VarPro), a method originally designed for separable nonlinear least-squares problems. Our main contribution is the Gauss--Newton VarPro method (GNvpro) that extends the reach of the VarPro idea to nonquadratic objective functions, most notably cross-entropy loss functions arising in classification. These extensions make GNvpro applicable to all training problems that involve a DNN whose last layer is an affine mapping, which is common in many state-of-the-art architectures. In our four numerical experiments from surrogate modeling, segmentation, and classification, GNvpro solves the optimization problem more efficiently than commonly used stochastic gradient descent (SGD) schemes. Finally, GNvpro finds solutions that generalize well, and in all but one example better than well-tuned SGD methods, to unseen data points.

97 MATHEMATICS AND COMPUTING↗

Estuarine Dissolved Organic Carbon Flux From Space: With Application to Chesapeake and Delaware Bays

This study uses a neural network model trained with in situ data, combined with satellite data and hydrodynamic model products, to compute the daily estuarine export of dissolved organic carbon (DOC) at the mouths of Chesapeake Bay (CB) and Delaware Bay (DB) from 2007 to 2011. Both bays show large flux variability with highest fluxes in spring and lowest in fall as well as interannual flux variability (0.18 and 0.27 Tg C/year in 2008 and 2010 for CB; 0.04 and 0.09 Tg C/year in 2008 and 2011 for DB). Based on previous estimates of total organic carbon (TOCexp) exported by all Mid-Atlantic Bight estuaries (1.2 Tg C/year), the DOC export (CB + DB) of 0.3 Tg C/year estimated here corresponds to 25% of the TOCexp. Spatial and temporal covariations of velocity and DOC concentration provide contributions to the flux, with larger spatial influence. Differences in the discharge of fresh water into the bays (74 billion cu. m/year for CB and 21 billion cu. m/year for DB) and their geomorphologies are major drivers of the differences in DOC fluxes for these two systems. Terrestrial DOC inputs are similar to the export of DOC at the bay mouths at annual and longer time scales but diverge significantly at shorter time scales (days to months). Future efforts will expand to the Mid-Atlantic Bight and Gulf of Maine, and its major rivers and estuaries, in combination with coupled terrestrial-estuarine-ocean biogeochemical models that include effects of climate change, such as warming and CO2 increase.

Estuarine DOC export↗

PINN surrogate of Li-ion battery models for parameter inference, Part II: Regularization and application of the pseudo-2D model

Bayesian parameter inference is useful to improve Li-ion battery diagnostics and can help formulate battery aging models. However, it is computationally intensive and cannot be easily repeated for multiple cycles, multiple operating conditions, or multiple replicate cells. To reduce the computational cost of Bayesian calibration, numerical solvers for physics-based models can be replaced with faster surrogates. A physics-informed neural network (PINN) is developed as a surrogate for the pseudo-2D (P2D) battery model calibration. For the P2D surrogate, additional training regularization was needed as compared to the PINN single-particle model (SPM) developed in Part I. Both the PINN SPM and P2D surrogate models are exercised for parameter inference and compared to data obtained from a direct numerical solution of the governing equations. A parameter inference study highlights the ability to use these PINNs to calibrate scaling parameters for the cathode Li diffusion and the anode exchange current density. By realizing computational speed-ups of ~2250x for the P2D model, as compared to using standard integrating methods, the PINN surrogates enable rapid state-of-health diagnostics. Finally, in the low-data availability scenario, the testing error was estimated to ~2 mV for the SPM surrogate and ~10 mV for the P2D surrogate which could be mitigated with additional data.

25 ENERGY STORAGE↗

Towards fast, accurate predictions of RF simulations via data-driven modeling: Forward and lateral models

Three machine learning techniques (multilayer perceptron, random forest, and Gaussian process) provide fast surrogate models for lower hybrid current drive (LHCD) simulations. A single GENRAY/CQL3D simulation without radial diffusion of fast electrons requires several minutes of wall-clock time to complete, which is acceptable for many purposes, but too slow for integrated modeling and real-time control applications. More accurate simulations with fast electron diffusion are even slower, requiring multiple hours of run time with parallel processing. The machine learning models use a database of 16,000+ GEN-RAY/CQL3D simulations for training, validation, and testing. Latin hypercube sampling methods implemented in πScope ensure that the database covers the range of 9 input parameters (n e0 , T e0 , I p , B t , R 0 , n ∥︀ , Z e f f , V loop , P LHCD ) with sufficient density in all regions of parameter space. The surrogate models reduce the computation time from minutes-hours to ms with high accuracy across the input parameter space. Data-driven surrogate models also allow for solving inverse and “lateral” problems. A surrogate model for the inverse problem maps from a desired current drive or power deposition profile to a set of input parameters that would result in such a profile, while a surrogate model for the lateral problem maps from a measured experimental quantity such as hard x-ray emission to a current drive or power deposition profile. In conclusion, the πScope database creation workflow is flexible and applicable to other RF simulation codes such as TORIC.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Reconstructing the exit wave of 2D materials in high-resolution transmission electron microscopy using machine learning

Reconstruction of the exit wave function is an important route to interpreting high-resolution transmission electron microscopy (HRTEM) images. Here we demonstrate that convolutional neural networks can be used to reconstruct the exit wave from a short focal series of HRTEM images, with a fidelity comparable to conventional exit wave reconstruction. We use a fully convolutional neural network based on the U-Net architecture, and demonstrate that we can train it on simulated exit waves and simulated HRTEM images of graphene-supported molybdenum disulphide (an industrial desulfurization catalyst). We then apply the trained network to analyse experimentally obtained images from similar samples, and obtain exit waves that clearly show the atomically resolved structure of both the MoS 2 nanoparticles and the graphene support. We also show that it is possible to successfully train the neural networks to reconstruct exit waves for 3400 different two-dimensional materials taken from the Computational 2D Materials Database of known and proposed two-dimensional materials.

2D materials↗

QC-GN 2 oMS 2 : a Graph Neural Net for High Resolution Mass Spectra Prediction

Predicting the mass spectrum of a molecular ion is often accomplished via three generalized approaches: rules-based methods for bond breaking, deep learning, or quantum chemical (QC) modeling. Rules-based approaches are often limited by the conditions for different chemical subspaces and perform poorly under chemical regimes with few defined rules. QC modeling is theoretically robust but requires significant amounts of computational time to produce a spectrum for a given target. Among deep learning techniques, graph neural networks (GNNs) have performed better than previous work with fingerprint-based neural networks in mass spectra prediction. To explore this technique further, we investigate the effects of including quantum chemically derived information as edge features in the GNN to increase predictive accuracy. The models we investigated include categorical bond order, bond force constants derived from extended tight-binding (xTB) quantum chemistry, and acyclic bond dissociation energies. Throughout this work, we evaluated these models against a control GNN with no edge features in the input graphs. Bond dissociation enthalpies yielded the best improvement with a cosine similarity score of 0.462 relative to the baseline model (0.437). In this work we also apply dynamic graph attention which improves performance on benchmark problems and supports the inclusion of edge features. Between implementations, we investigate the nature of the molecular embedding for spectra prediction and discuss the recognition of fragment topographies in distinct chemistries for further development in tandem mass spectrometry prediction.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Electron-phonon interaction and thermal boundary resistance at the crystal-amorphous interface of the phase change compound GeTe

Phonon dispersion relations and electron-phonon coupling of hole-doped trigonal GeTe have been computed by density functional perturbation theory. This compound is a prototypical phase change material of interest for applications in phase change non-volatile memories. The calculations allowed us to estimate the electron-phonon contribution to the thermal boundary resistance at the interface between the crystalline and amorphous phases present in the device. The lattice contribution to the thermal boundary resistance has been computed by non-equilibrium molecular dynamics simulations with an interatomic potential based on a neural network scheme. We find that the electron-phonon term contributes to the thermal boundary resistance to an extent which is strongly dependent on the concentration and mobility of the holes. Further, for measured values of the holes concentration and electrical conductivity, the electron-phonon term is larger than the contribution from the lattice. It is also shown that the presence of Ge vacancies, responsible for the p-type degenerate character of the semiconductor, strongly affects the lattice thermal conductivity of the crystal.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

(Bi 0.2 Sb 0.8 ) 2 Te 3 based dynamic synapses with programmable spatio-temporal dynamics

Neuromorphic computing has recently emerged as a promising paradigm to overcome the von-Neumann bottleneck and enable orders of magnitude improvement in bandwidth and energy efficiency. However, existing complementary metal-oxide-semiconductor (CMOS) digital devices, the building block of our computing system, are fundamentally different from the analog synapses, the building block of the biological neural network—rendering the hardware implementation of the artificial neural networks (ANNs) not scalable in terms of area and power, with existing CMOS devices. In addition, the spatiotemporal dynamic, a crucial component for cognitive functions in the neural network, has been difficult to replicate with CMOS devices. Here, we present the first topological insulator (TI) based electrochemical synapse with programmable spatiotemporal dynamics, where long-term and short-term plasticity in the TI synapse are achieved through the charge transfer doping and ionic gating effects, respectively. We also demonstrate basic neuronal functions such as potentiation/depression and paired-pulse facilitation with high precision (>500 states per device), as well as a linear and symmetric weight update. We envision that the dynamic TI synapse, which shows promising scaling potential in terms of energy and speed, can lead to the hardware acceleration of truly neurorealistic ANNs with superior cognitive capabilities and excellent energy efficiency.

36 MATERIALS SCIENCE↗

A Visual Designer of Layer‐wise Relevance Propagation Models

Abstract Layer‐wise Relevance Propagation (LRP) is an emerging and widely‐used method for interpreting the prediction results of convolutional neural networks (CNN). LRP developers often select and employ different relevance backpropagation rules and parameters, to compute relevance scores on input images. However, there exists no obvious solution to define a “best” LRP model. A satisfied model is highly reliant on pertinent images and designers' goals. We develop a visual model designer, named as VisLRPDesigner, to overcome the challenges in the design and use of LRP models. Various LRP rules are unified into an integrated framework with an intuitive workflow of parameter setup. VisLRPDesigner thus allows users to interactively configure and compare LRP models. It also facilitates relevance‐based visual analysis with two important functions: relevance‐based pixel flipping and neuron ablation. Several use cases illustrate the benefits of VisLRPDesigner. The usability and limitation of the visual designer is evaluated by LRP users.

Huang, Xinyi↗

BSTC: A Novel Binarized-Soft-Tensor-Core Design for Accelerating Bit-Based Approximated Neural Nets

Binarized neural network (or BNN) promises tremendous performance improvement over traditional DNNs through simplified bit-level computation and significantly reduced memory access/storage cost. In addition, it has other advantages of low-cost, low-energy and high-robustness, showing great utilization potential in resources-constrained, volatile and latency-critical applications, which are critical for future HPC execution. However, the promised significant performance gain of BNN inference has never been fully demonstrated on general-purpose processors, particularly on GPUs, due to: (i) the challenge to extract and leverage sufficient fine-grained bit-level-parallelism to saturate GPU cores when batch is small; (ii) the fundamental design conflict between bit-based BNN algorithm and underlying word-based architecture; (iii) architecture \& performance unfriendly BNN network design. To address (i) and (ii), we propose binarized-soft-tensor-core as a software-hardware codesign approach to construct bit-manipulation capability for modern GPUs to effectively harvest the emerging bit-level-parallelism. To tackle (iii), we propose intra- and inter-layer fusion techniques so that the entire BNN inference can be packed into a single GPU kernel, to avoid high-cost frequent launching. Experiments demonstrate that our design can achieve over 1000x speedup for raw inference latency and 10x for inference throughput over the state-of-the-art full-precision simulated BNN inference for AlexNet on ImageNet.

Li, Ang↗

Scaling Inference Using Triton to Accelerate Particle Physics at the LHC and DUNE

DUNE and the LHC experiments consist of unique and cutting-edge particle detectors that create massive, complex, and rich datasets with billions of events. They require sophisticated algorithms to reconstruct and interpret the data. Modern machine learning algorithms provide a powerful toolset to detect and classify particles, from familiar image processing convolutional neural networks to newer graph neural network architectures. A full reconstruction of these particle collisions requires novel approaches to handle the computing challenge of processing so much raw data. In a series of studies, physicists from Fermilab, CERN, and university groups explored how to accelerate their data processing using the Triton Inference Server.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Quantum Memristors in Frequency-Entangled Optical Fields

A quantum memristor is a passive resistive circuit element with memory, engineered in a given quantum platform. It can be represented by a quantum system coupled to a dissipative environment, in which a system–bath coupling is mediated through a weak measurement scheme and classical feedback on the system. In quantum photonics, such a device can be designed from a beam splitter with tunable reflectivity, which is modified depending on the results of measurements in one of the outgoing beams. Here, we show that a similar implementation can be achieved with frequency-entangled optical fields and a frequency mixer that, working similarly to a beam splitter, produces state superpositions. We show that the characteristic hysteretic behavior of memristors can be reproduced when analyzing the response of the system with respect to the control, for different experimentally attainable states. Since memory effects in memristors can be exploited for classical and neuromorphic computation, the results presented in this work could be a building block for constructing quantum neural networks in quantum photonics, when scaling up.

36 MATERIALS SCIENCE↗

Real-time neuromorphic algorithms for inverse kinematics of redundant manipulators

The paper presents an efficient neuromorphic formulation to accurately solve the inverse kinematics problem for redundant manipulators. The approach involves a dynamical learning procedure based on a novel formalism in neural network theory: the concept of 'terminal' attractors. Topographically mapped terminal attractors are used to define a neural network whose synaptic elements can rapidly encapture the inverse kinematics transformations, and, subsequently generalize to compute joint-space coordinates required to achieve arbitrary end-effector configurations. Unlike prior neuromorphic implementations, this technique can also systematically exploit redundancy to optimize kinematic criteria, e.g., torque optimization. Simulations on 3-DOF and 7-DOF redundant manipulators, are used to validate the theoretical framework and illustrate its computational efficacy.

Barhen, Jacob↗

The TENNLab Suite of LIDAR-Based Control Applications for Recurrent, Spiking, Neuromorphic Systems

Recurrent, spiking neuromorphic systems (RSNS’s) have several desirable features compared to conventional neural networks when applied to applications that require real-time control. The TENNLab Exploratory Neuromorphic Computing Framework contains a variety of real-time control applications that use LIDAR for input, and allow a trained RSNS to make real-time decisions for navigation, surveillance and targeting. While each application is interesting on its own, the collection of applications is very useful for benchmarking RSNS’ s, evaluating training algorithms, and designing neuromorphic processing elements. In this paper, we describe the applications that compose the suite, and use them to benchmark four different RSNS’s.

Plank, James↗

A Visual Designer of Layer-wise Relevance Propagation Models

Layer-wise Relevance Propagation (LRP) is an emerging and widely-used method for interpreting the prediction results of convolutional neural networks (CNN). LRP developers often select and employ different relevance backpropagation rules and parameters, to compute relevance scores on input images. However, there exists no obvious solution to define a "best" LRP model. A satisfied model is highly reliant on pertinent images and designers' goals. We develop a visual model designer, named as VisLRPDesigner, to overcome the challenges in the design and use of LRP models. Various LRP rules are unified into an integrated framework with an intuitive workflow of parameter setup. VisLRPDesigner thus allows users to interactively configure and compare LRP models. It also facilitates relevance-based visual analysis with two important functions: relevance-based pixel flipping and neuron ablation. Several use cases illustrate the benefits of VisLRPDesigner. The usability and limitation of the visual designer is evaluated by LRP users.

97 MATHEMATICS AND COMPUTING↗