Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Neural network”

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 361 records · Page 20
All sourcesNASA NTRSDOE OSTI
NASA NTRS1995

Experiments in Neural-Network Control of a Free-Flying Space Robot

Four important generic issues are identified and addressed in some depth in this thesis as part of the development of an adaptive neural network based control system for an experimental free flying space robot prototype. The first issue concerns the importance of true system level design of the control system. A new hybrid strategy is developed here, in depth, for the beneficial integration of neural networks into the total control system. A second important issue in neural network control concerns incorporating a priori knowledge into the neural network. In many applications, it is possible to get a reasonably accurate controller using conventional means. If this prior information is used purposefully to provide a starting point for the optimizing capabilities of the neural network, it can provide much faster initial learning. In a step towards addressing this issue, a new generic Fully Connected Architecture (FCA) is developed for use with backpropagation. A third issue is that neural networks are commonly trained using a gradient based optimization method such as backpropagation; but many real world systems have Discrete Valued Functions (DVFs) that do not permit gradient based optimization. One example is the on-off thrusters that are common on spacecraft. A new technique is developed here that now extends backpropagation learning for use with DVFs. The fourth issue is that the speed of adaptation is often a limiting factor in the implementation of a neural network control system. This issue has been strongly resolved in the research by drawing on the above new contributions.

NEURAL NETS↗
DOE OSTI2024

Making Phase-Picking Neural Networks More Consistent and Interpretable

Improving the interpretability of phase-picking neural networks remains an important task to facilitate their deployment to routine, real-time seismic monitoring. The popular phase-picking neural networks published in the literature lack interpretability because their output prediction scores do not necessarily correspond with the reliability of phase picks and can even be highly inconsistent depending on how we window the waveform data. Here, we show that systematically shifting the waveforms during training and using an antialiasing filter within the neural network architecture can substantially improve the consistency of the output prediction scores and can even make them scale with the signal-to-noise ratios of the waveforms. We demonstrate the improvements by applying these approaches to a commonly used phase-picking neural network architecture and using waveform data from the 2019 Ridgecrest earthquake sequence.

58 GEOSCIENCES↗
DOE OSTI2022

Towards a robust out-of-the-box neural network model for genomic data

The accurate prediction of biological features from genomic data is paramount for precision medicine and sustainable agriculture. For decades, neural network models have been widely popular in fields like computer vision, astrophysics and targeted marketing given their prediction accuracy and their robust performance under big data settings. Yet neural network models have not made a successful transition into the medical and biological world due to the ubiquitous characteristics of biological data such as modest sample sizes, sparsity, and extreme heterogeneity. Here, we investigate the robustness, generalization potential and prediction accuracy of widely used convolutional neural network and natural language processing models with a variety of heterogeneous genomic datasets. Mainly, recurrent neural network models outperform convolutional neural network models in terms of prediction accuracy, overfitting and transferability across the datasets under study. While the perspective of a robust out-of-the-box neural network model is out of reach, we identify certain model characteristics that translate well across datasets and could serve as a baseline model for translational researchers.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRS1988

Hierarchical neural networks for autonomous data analysis and decision making

A neural network based data analysis and decision making system to increase the autonomy of a planetary rover or similar exploratory vehicle is presented. A hierarchical series of neural networks for real time analysis of scientific images is used. The system under development emphasizes analysis of multispectral images by classifier and feature detector neural networks, to provide information on the mineral composition of a scene. A hierarchy of alternating analysis and decision making networks is being developed to allow increasingly fine scale analysis in regions of the image that are potentially important. It is noted that this system will facilitate both the selection of high priorty scientific information for transmission to earth, and the autonomous collection of rocks and soil for sample return.

Eberlein, Susan↗
DOE OSTI2023

Analytic Theory for the Dynamics of Wide Quantum Neural Networks

Here, parametrized quantum circuits can be used as quantum neural networks and have the potential to outperform their classical counterparts when trained for addressing learning problems. To date, much of the results on their performance on practical problems are heuristic in nature. In particular, the convergence rate for the training of quantum neural networks is not fully understood. Here, we analyze the dynamics of gradient descent for the training error of a class of variational quantum machine learning models. We define wide quantum neural networks as parametrized quantum circuits in the limit of a large number of qubits and variational parameters. Then, we find a simple analytic formula that captures the average behavior of their loss function and discuss the consequences of our findings. For example, for random quantum circuits, we predict and characterize an exponential decay of the residual training error as a function of the parameters of the system. Finally, we validate our analytic results with numerical experiments.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
NASA NTRS1991

Learning control of inverted pendulum system by neural network driven fuzzy reasoning: The learning function of NN-driven fuzzy reasoning under changes of reasoning environment

Whereas conventional fuzzy reasonings are associated with tuning problems, which are lack of membership functions and inference rule designs, a neural network driven fuzzy reasoning (NDF) capable of determining membership functions by neural network is formulated. In the antecedent parts of the neural network driven fuzzy reasoning, the optimum membership function is determined by a neural network, while in the consequent parts, an amount of control for each rule is determined by other plural neural networks. By introducing an algorithm of neural network driven fuzzy reasoning, inference rules for making a pendulum stand up from its lowest suspended point are determined for verifying the usefulness of the algorithm.

Hayashi, Isao↗
NASA NTRS1990

Neural networks and MIMD-multiprocessors

Two artificial neural network models are compared. They are the Hopfield Neural Network Model and the Sparse Distributed Memory model. Distributed algorithms for both of them are designed and implemented. The run time characteristics of the algorithms are analyzed theoretically and tested in practice. The storage capacities of the networks are compared. Implementations are done using a distributed multiprocessor system.

Vanhala, Jukka↗
NASA NTRS1990

Neural network approach to Space Shuttle Main Engine health monitoring

A neural network was trained to distinguish anomalies in Space Shuttle Main Engine sensor data from noisy normal steady-state sensor data. Power spectra of successive windows of individual sensor data were presented to a neural network using Kohonen's topological feature map training algorithm. The trained network for each sensor was then tested to determine if it would detect anomalies in the sensor data, and if so, the time at which the anomaly would be detected. Power spectra from a few hundred seconds of actual test data from NASA tests 901-364 and 904-044 were used to test the network. In both cases, the neural network detected the onset of anomalous engine behavior at approximately the same time within each test as the onset times reported by NASA and Rocketdyne experts in their post-test analyses.

Whitehead, Bruce A.↗
DOE OSTI2022

Towards Compact Neural Networks via End-to-End Training: A Bayesian Tensor Approach with Automatic Rank Determination

Post-training model compression can reduce the inference costs of deep neural networks, but uncompressed training still consumes enormous hardware resources and energy. To enable low-energy training on edge devices, it is highly desirable to directly train a compact neural network from scratch with a low memory cost. Low-rank tensor decomposition is an effective approach to reduce the memory and computing costs of large neural networks. However, directly training low-rank tensorized neural networks is a very challenging task because it is hard to determine a proper tensor rank a priori, and the tensor rank controls both model complexity and accuracy. Here, this paper presents a novel end-to-end framework for low-rank tensorized training. We first develop a Bayesian model that supports various low-rank tensor formats (e.g., CANDECOMP/PARAFAC, Tucker, tensor-train, and tensor-train matrix) and reduces neural network parameters with automatic rank determination during training. Then we develop a customized Bayesian solver to train large-scale tensorized neural networks. Our training methods shows orders-of-magnitude parameter reduction and little accuracy loss (or even better accuracy) in the experiments. On a very large deep learning recommendation system with over 4.2 ×10 9 model parameters, our method can reduce the parameter number to 1.6 ×10 5 automatically in the training process (i.e., by 2.6 ×10 4 times) while achieving almost the same accuracy. Code is available at https://github.com/colehawkins/bayesian-tensor-rank-determination.

compact neural networks↗
NASA NTRS1994

Comparisons of neural networks to standard techniques for image classification and correlation

Neural network techniques for multispectral image classification and spatial pattern detection are compared to the standard techniques of maximum-likelihood classification and spatial correlation. The neural network produced a more accurate classification than maximum-likelihood of a Landsat scene of Tucson, Arizona. Some of the errors in the maximum-likelihood classification are illustrated using decision region and class probability density plots. As expected, the main drawback to the neural network method is the long time required for the training stage. The network was trained using several different hidden layer sizes to optimize both the classification accuracy and training speed, and it was found that one node per class was optimal. The performance improved when 3x3 local windows of image data were entered into the net. This modification introduces texture into the classification without explicit calculation of a texture measure. Larger windows were successfully used for the detection of spatial features in Landsat and Magellan synthetic aperture radar imagery.

Paola, Justin D.↗
DOE OSTI2022

GAAF: Searching Activation Functions for Binary Neural Networks Through Genetic Algorithm

Binary neural networks (BNNs) show promising utilization in cost and power-restricted domains such as edge devices and mobile systems. This is due to its significantly less computation and storage demand, but at the cost of degraded performance. To close the accuracy gap, in this paper we propose to add a complementary activation function (AF) ahead of the sign based binarization, and rely on the genetic algorithm (GA) to automatically search for the ideal AFs. These AFs can help extract extra information from the input data in the forward pass, while allowing improved gradient approximation in the backward pass. Fifteen novel AFs are identified through our GA-based search, while most of them show improved performance (up to 2.54% on ImageNet) when testing on different datasets and network models. Interestingly, periodic functions are identified as a key component for most of the discovered AFs, which rarely exist in human designed AFs. Our method offers a novel approach for designing general and application-specific BNN architecture.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRS1993

Application of neural network to remote sensing of soil moisture using theoretical polarimetric backscattering coefficients

This paper investigates the potential application of neural network to inversion of soil moisture using polarimetric remote sensing data. The neural network used for the inversion of soil parameters is multi-layer perceptron trained with the back-propagation algorithm. The training data include the polarimetric backscattering coefficients obtained from theoretical surface scattering models together with an assumed nominal range of soil parameters which are comprised of the soil permittivity and surface roughness parameters. Soil permittivity is calculated from the soil moisture and the assumed soil texture based on an empirical formula at C-, L-, and P-bands. The rough surface parameters for the soil surface, which is described by the Gaussian random process, are the root-mean-square (rms) height and correlation length. For the rough surface scattering, small perturbation method is used for the L-band frequency, and Kirchhoff approximation is used for the C-band frequency to obtain the corresponding backscattering coefficients. During the training, the backscattering coefficients are the inputs to the neural net and the output from the net are compared with the desired soil parameters to adjust the interconnecting weights. The process is repeated for each input-output data entry and then for the entire training data until convergence is reached. After training, the backscattering coefficients are applied to the trained neural net to retrieve the soil parameters which are compared with the desired soil parameters to verify the effectiveness of this technique. Several cases are examined. First, for simplicity, the correlation length and rms height of the soil surface are fixed while soil moisture is varied. Soil moisture obtained using the neural networks with either L-band or C-band backscattering coefficients for the HH and VV polarizations as inputs is in good agreement with the desired soil moisture. The neural net output matches the desired output for the soil moisture range of 16 to 60 percent for the C-band case. The next case investigated is to vary both soil moisture and rms height while keeping the correlation length fixed. For this case, C-band backscattering coefficients are not sufficient for retrieving two parameters because the Kirchhoff approximation gives the same HH and VV backscattering coefficients. Therefore, the backscattering coefficients at two different frequency bands are necessary to find both the soil moisture and rms height. Finally, the neural nets are also applied to simultaneously invert soil moisture, rms height, and correlation length. Overall, the soil moisture retrieved from the neural network agrees very well with the desired soil moisture. This suggests that the neural network shows potential for retrieval of soil parameters from remote sensing data.

Wang, L.↗
NASA NTRS1990

Statistical methods and neural network approaches for classification of data from multiple sources

Statistical methods for classification of data from multiple data sources are investigated and compared to neural network models. A problem with using conventional multivariate statistical approaches for classification of data of multiple types is in general that a multivariate distribution cannot be assumed for the classes in the data sources. Another common problem with statistical classification methods is that the data sources are not equally reliable. This means that the data sources need to be weighted according to their reliability but most statistical classification methods do not have a mechanism for this. This research focuses on statistical methods which can overcome these problems: a method of statistical multisource analysis and consensus theory. Reliability measures for weighting the data sources in these methods are suggested and investigated. Secondly, this research focuses on neural network models. The neural networks are distribution free since no prior knowledge of the statistical distribution of the data is needed. This is an obvious advantage over most statistical classification methods. The neural networks also automatically take care of the problem involving how much weight each data source should have. On the other hand, their training process is iterative and can take a very long time. Methods to speed up the training procedure are introduced and investigated. Experimental results of classification using both neural network models and statistical methods are given, and the approaches are compared based on these results.

Benediktsson, Jon Atli↗
DOE OSTI2020

Atomic resolution convergent beam electron diffraction analysis using convolutional neural networks

Two types of convolutional neural network (CNN) models, a discrete classification network and a continuous regression network, were trained to determine local sample thickness from convergent beam diffraction (CBED) patterns of SrTiO 3 collected in a scanning transmission electron microscope (STEM) at atomic column resolution. Acquisition of atomic resolution CBED patterns for this purpose requires careful balancing of CBED feature size in pixels, acquisition speed, and detector dynamic range. The training datasets were derived from multislice simulations, which must be convolved with incoherent source broadening. Sample thicknesses were also determined using quantitative high-angle annular dark-field (HAADF) STEM images acquired simultaneously. The regression CNN performed well on sample thinner than 35 nm, with 70% of the CNN results within 1 nm of HAADF thickness, and 1.0 nm overall root mean square error between the two measurements. The classification CNN was trained for a thicknesses up to 100 nm and yielded 66% of CNN results within one classification increment of 2 nm of HAADF thickness. Our approach depends on methods from computer vision including transfer learning and image augmentation.

36 MATERIALS SCIENCE↗
NASA NTRS1992

Decision boundary feature extraction for neural networks

We propose a new feature extraction method for neural networks. The method is based on the recently published decision boundary feature extraction algorithm. It has been shown that all the necessary features for classification can be extracted from the decision boundary. To apply the decision boundary feature extraction method, we first define the decision boundary in neural networks. Next, we propose a procedure for extracting all the necessary features for classification from the decision boundary. The proposed algorithm preserves the characteristics of neural networks, which can define arbitrary decision boundary. Experiments show promising results.

Lee, Chulhee↗
DOE OSTI2021

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↗
DOE OSTI2019

Application of deep artificial neural networks to multi-dimensional flamelet libraries and spray flames

The “curse of dimensionality” has limited the applicability and expansion of tabulated combustion models. While the tabulated flamelet model and other multi-dimensional manifold approaches have shown predictive capability, the associated tabulation involves the storage of large lookup tables, requiring large memory as well as multi-dimensional interpolation subroutines, all implemented during runtime. As such, this work investigates the use of deep artificial neural networks to replace lookup tables in order to reduce the memory footprint and increase the computational speed of tabulated flamelets and related approaches. Specifically, different strategic approaches to training the artificial neural network models are explored and a grouped multi-target artificial neural network is introduced, which takes advantage of the ability of artificial neural networks to map an input space to multiple targets by classifying the species based on their correlation to one another. The grouped multi-target artificial neural network approach is validated by applying it to an n-dodecane spray flame using conditions of the Spray A flame from the Engine Combustion Network and comparing global flame characteristics for different ambient conditions using a well-established large-eddy simulation framework. The same framework is then extended to the simulations of methyl decanoate combustion in a compression ignition engine. The validation studies show that the grouped multi-target artificial neural networks are able to accurately capture flame liftoff, autoignition, two-stage heat release and other quantitative trends over a range of conditions. The use of neural networks in conjunction with the grouping mechanism as performed in the grouped multi-target artificial neural network produces a significant reduction in the memory footprint and computational costs for the code and, thus, widens the operating envelope for higher fidelity engine simulations with detailed mechanisms.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1999

Predicting Slag Generation in Sub-Scale Test Motors Using a Neural Network

Generation of slag (aluminum oxide) is an important issue for the Reusable Solid Rocket Motor (RSRM). Thiokol performed testing to quantify the relationship between raw material variations and slag generation in solid propellants by testing sub-scale motors cast with propellant containing various combinations of aluminum fuel and ammonium perchlorate (AP) oxidizer particle sizes. The test data were analyzed using statistical methods and an artificial neural network. This paper primarily addresses the neural network results with some comparisons to the statistical results. The neural network showed that the particle sizes of both the aluminum and unground AP have a measurable effect on slag generation. The neural network analysis showed that aluminum particle size is the dominant driver in slag generation, about 40% more influential than AP. The network predictions of the amount of slag produced during firing of sub-scale motors were 16% better than the predictions of a statistically derived empirical equation. Another neural network successfully characterized the slag generated during full-scale motor tests. The success is attributable to the ability of neural networks to characterize multiple complex factors including interactions that affect slag generation.

Wiesenberg, Brent↗