Search NASA⌕ Search

SEARCH · Search NASA

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 217 records · Page 12

HGQ: High Granularity Quantization for Real-time Neural Networks on FPGAs

Neural networks with sub-microsecond inference latency are required by many critical applications. Targeting such applications deployed on FPGAs, we present High Granularity Quantization (HGQ), a quantization-aware training framework that optimizes parameter bit-widths through gradient descent. Unlike conventional methods, HGQ determines the optimal bit-width for each parameter independently, making it suitable for hardware platforms supporting heterogeneous arbitrary precision arithmetic. In our experiments, HGQ shows superior performance compared to existing network compression methods, achieving orders of magnitude reduction in resource consumption and latency while maintaining the accuracy on several benchmark tasks. These improvements enable the deployment of complex models previously infeasible due to resource or latency constraints. HGQ is open-source and is used for developing next-generation trigger systems at the CERN ATLAS and CMS experiments for particle physics, enabling the use of advanced machine learning models for real-time data selection with sub-microsecond latency.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Orthogonal patterns in binary neural networks

A binary neural network that stores only mutually orthogonal patterns is shown to converge, when probed by any pattern, to a pattern in the memory space, i.e., the space spanned by the stored patterns. The latter are shown to be the only members of the memory space under a certain coding condition, which allows maximum storage of M=(2N) sup 0.5 patterns, where N is the number of neurons. The stored patterns are shown to have basins of attraction of radius N/(2M), within which errors are corrected with probability 1 in a single update cycle. When the probe falls outside these regions, the error correction capability can still be increased to 1 by repeatedly running the network with the same probe.

Baram, Yoram↗

The performance evaluation of a new neural network based traffic management scheme for a satellite communication network

A neural-network-based traffic management scheme for a satellite communication network is described. The scheme consists of two levels of management. The front end of the scheme is a derivation of Kohonen's self-organization model to configure maps for the satellite communication network dynamically. The model consists of three stages. The first stage is the pattern recognition task, in which an exemplar map that best meets the current network requirements is selected. The second stage is the analysis of the discrepancy between the chosen exemplar map and the state of the network, and the adaptive modification of the chosen exemplar map to conform closely to the network requirement (input data pattern) by means of Kohonen's self-organization. On the basis of certain performance criteria, whether a new map is generated to replace the original chosen map is decided in the third stage. A state-dependent routing algorithm, which arranges the incoming call to some proper path, is used to make the network more efficient and to lower the call block rate. Simulation results demonstrate that the scheme, which combines self-organization and the state-dependent routing mechanism, provides better performance in terms of call block rate than schemes that only have either the self-organization mechanism or the routing mechanism.

Ansari, Nirwan↗

Material Data Representation of Hysteresis Loops for Hastelloy X Using Artificial Neural Networks

The artificial neural network (ANN) model proposed by Rumelhart, Hinton, and Williams is applied to develop a functional approximation of material data in the form of hysteresis loops from a nickel-base superalloy, Hastelloy X. Several different ANN configurations are used to model hysteresis loops at different cycles for this alloy. The ANN models were successful in reproducing the hysteresis loops used for its training. However, because of sharp bends at the two ends of hysteresis loops, a drift occurs at the corners of the loops where loading changes to unloading and vice versa (the sharp bends occurred when the stress-strain curves were reproduced by adding stress increments to the preceding values of the stresses). Therefore, it is possible only to reproduce half of the loading path. The generalization capability of the network was tested by using additional data for two other hysteresis loops at different cycles. The results were in good agreement. Also, the use of ANN led to a data compression ratio of approximately 22:1.

Alam, Javed↗

Verification and Validation of KBS with Neural Network Components

Artificial Neural Network (ANN) play an important role in developing robust Knowledge Based Systems (KBS). The ANN based components used in these systems learn to give appropriate predictions through training with correct input-output data patterns. Unlike traditional KBS that depends on a rule database and a production engine, the ANN based system mimics the decisions of an expert without specifically formulating the if-than type of rules. In fact, the ANNs demonstrate their superiority when such if-then type of rules are hard to generate by human expert. Verification of traditional knowledge based system is based on the proof of consistency and completeness of the rule knowledge base and correctness of the production engine.These techniques, however, can not be directly applied to ANN based components.In this position paper, we propose a verification and validation procedure for KBS with ANN based components. The essence of the procedure is to obtain an accurate system specification through incremental modification of the specifications using an ANN rule extraction algorithm.

Wen, Wu↗

Estimating Surface Soil Moisture from SMAP Observations Using a Neural Network Technique

A Neural Network (NN) algorithm was developed to estimate global surface soil moisture for April 2015 to March 2017 with a 2-3 day repeat frequency using passive microwave observations from the Soil Moisture Active Passive (SMAP) satellite, surface soil temperatures from the NASA Goddard Earth Observing System Model version 5 (GEOS-5) land modeling system, and Moderate Resolution Imaging Spectroradiometer-based vegetation water content. The NN was trained on GEOS-5 soil moisture target data, making the NN estimates consistent with the GEOS-5 climatology, such that they may ultimately be assimilated into this model without further bias correction. Evaluated against in situ soil moisture measurements, the average unbiased root mean square error (ubRMSE), correlation and anomaly correlation of the NN retrievals were 0.037 m(exp. 3)m(exp. -3), 0.70 and 0.66, respectively, against SMAP core validation site measurements and 0.026 m(exp. 3)m(exp. -3), 0.58 and 0.48, respectively, against International Soil Moisture Network (ISMN) measurements. At the core validation sites, the NN retrievals have a significantly higher skill than the GEOS-5 model estimates and a slightly lower correlation skill than the SMAP Level-2 Passive (L2P) product. The feasibility of the NN method was reflected by a lower ubRMSE compared to the L2P retrievals as well as a higher skill when ancillary parameters in physically-based retrievals were uncertain. Against ISMN measurements, the skill of the two retrieval products was more comparable. A triple collocation analysis against Advanced Microwave Scanning Radiometer 2 (AMSR2) and Advanced Scatterometer (ASCAT) soil moisture retrievals showed that the NN and L2P retrieval errors have a similar spatial distribution, but the NN retrieval errors are generally lower in densely vegetated regions and transition zones.

Soil Moisture Remote Sensing↗

Implementations of learning control systems using neural networks

The systematic storage in neural networks of prior information to be used in the design of various control subsystems is investigated. Assuming that the prior information is available in a certain form (namely, input/output data points and specifications between the data points), a particular neural network and a corresponding parameter design method are introduced. The proposed neural network addresses the issue of effectively using prior information in the areas of dynamical system (plant and controller) modeling, fault detection and identification, information extraction, and control law scheduling.

Sartori, Michael A.↗

Neural network applications in telecommunications

Neural network capabilities include automatic and organized handling of complex information, quick adaptation to continuously changing environments, nonlinear modeling, and parallel implementation. This viewgraph presentation presents Bellcore work on applications, learning chip computational function, learning system block diagram, neural network equalization, broadband access control, calling-card fraud detection, software reliability prediction, and conclusions.

Alspector, Joshua↗

A comparison of two neural network schemes for navigation

Neural networks have been applied to tasks in several areas of artificial intelligence, including vision, speech, and language. Relatively little work has been done in the area of problem solving. Two approaches to path-finding are presented, both using neural network techniques. Both techniques require a training period. Training under the back propagation (BPL) method was accomplished by presenting representations of (current position, goal position) pairs as input and appropriate actions as output. The Hebbian/interactive activation (HIA) method uses the Hebbian rule to associate points that are nearby. A path to a goal is found by activating a representation of the goal in the network and processing until the current position is activated above some threshold level. BPL, using back-propagation learning, failed to learn, except in a very trivial fashion, that is equivalent to table lookup techniques. HIA, performed much better, and required storage of fewer weights. In drawing a comparison, it is important to note that back propagation techniques depend critically upon the forms of representation used, and can be sensitive to parameters in the simulations; hence the BPL technique may yet yield strong results.

Munro, Paul W.↗

A comparison of two neural network schemes for navigation

Neural networks have been applied to tasks in several areas of artificial intelligence, including vision, speech, and language. Relatively little work has been done in the area of problem solving. Two approaches to path-finding are presented, both using neural network techniques. Both techniques require a training period. Training under the back propagation (BPL) method was accomplished by presenting representations of current position, goal position pairs as input and appropriate actions as output. The Hebbian/interactive activation (HIA) method uses the Hebbian rule to associate points that are nearby. A path to a goal is found by activating a representation of the goal in the network and processing until the current position is activated above some threshold level. BPL, using back-propagation learning, failed to learn, except in a very trivial fashion, that is equivalent to table lookup techniques. HIA, performed much better, and required storage of fewer weights. In drawing a comparison, it is important to note that back propagation techniques depend critically upon the forms of representation used, and can be sensitive to parameters in the simulations; hence the BPL technique may yet yield strong results.

Munro, Paul↗

Electronic neural networks for global optimization

An electronic neural network with feedback architecture, implemented in analog custom VLSI is described. Its application to problems of global optimization for dynamic assignment is discussed. The convergence properties of the neural network hardware are compared with computer simulation results. The neural network's ability to provide optimal or near optimal solutions within only a few neuron time constants, a speed enhancement of several orders of magnitude over conventional search methods, is demonstrated. The effect of noise on the circuit dynamics and the convergence behavior of the neural network hardware is also examined.

Thakoor, A. P.↗

Neural Networks for Rapid Design and Analysis

Artificial neural networks have been employed for rapid and efficient dynamics and control analysis of flexible systems. Specifically, feedforward neural networks are designed to approximate nonlinear dynamic components over prescribed input ranges, and are used in simulations as a means to speed up the overall time response analysis process. To capture the recursive nature of dynamic components with artificial neural networks, recurrent networks, which use state feedback with the appropriate number of time delays, as inputs to the networks, are employed. Once properly trained, neural networks can give very good approximations to nonlinear dynamic components, and by their judicious use in simulations, allow the analyst the potential to speed up the analysis process considerably. To illustrate this potential speed up, an existing simulation model of a spacecraft reaction wheel system is executed, first conventionally, and then with an artificial neural network in place.

Sparks, Dean W., Jr.↗

Survivability prediction of colon cancer patients using neural networks

We utilize deep neural networks to develop prediction models for patient survival and conditional survival of colon cancer. Our models are trained and validated on data obtained from the Surveillance, Epidemiology, and End Results Program. We provide an online outcome calculator for 1, 2, and 5 years survival periods. We experimented with multiple neural network structures and found that a network with five hidden layers produces the best results for these data. Moreover, the online outcome calculator provides conditional survival of 1, 2, and 5 years after surviving the mentioned survival periods. In this article, we report an approximate 0.87 area under the receiver operating characteristic curve measurements, higher than the 0.85 reported by Stojadinovic et al.

Health Care Sciences & Services↗

Quantifying the generalization error in deep learning in terms of data distribution and neural network smoothness

We report the accuracy of deep learning, i.e., deep neural networks, can be characterized by dividing the total error into three main types: approximation error, optimization error, and generalization error. Whereas there are some satisfactory answers to the problems of approximation and optimization, much less is known about the theory of generalization. Most existing theoretical works for generalization fail to explain the performance of neural networks in practice. To derive a meaningful bound, we study the generalization error of neural networks for classification problems in terms of data distribution and neural network smoothness. We introduce the cover complexity (CC) to measure the difficulty of learning a data set and the inverse of the modulus of continuity to quantify neural network smoothness. A quantitative bound for expected accuracy/error is derived by considering both the CC and neural network smoothness. Although most of the analysis is general and not specific to neural networks, we validate our theoretical assumptions and results numerically for neural networks by several data sets of images. The numerical results confirm that the expected error of trained networks scaled with the square root of the number of classes has a linear relationship with respect to the CC. We also observe a clear consistency between test loss and neural network smoothness during the training process. In addition, we demonstrate empirically that the neural network smoothness decreases when the network size increases whereas the smoothness is insensitive to training dataset size.

97 MATHEMATICS AND COMPUTING↗

Preconditioning electromyographic data for an upper extremity model using neural networks

A back propagation neural network has been employed to precondition the electromyographic signal (EMG) that drives a computational model of the human upper extremity. This model is used to determine the complex relationship between EMG and muscle activation, and generates an optimal muscle activation scheme that simulates the actual activation. While the experimental and model predicted results of the ballistic muscle movement are very similar, the activation function between the start and the finish is not. This neural network preconditions the signal in an attempt to more closely model the actual activation function over the entire course of the muscle movement.

Roberson, D. J.↗

The Value of Hyperparameter Optimization in Phase-Picking Neural Networks

The effectiveness of neural networks for picking seismic phase arrival times has been demonstrated through several case studies, and seismic monitoring programs are starting to adopt the technology into their workflows. However, published models were designed and trained using rather arbitrary choices of hyperparameters, limiting their performance. In this study, we use phase picks from both routine and template-matching analyses from multiple regions (Ridgecrest, California; Kilauea, Hawaii; Yellowstone, Wyoming–Montana–Idaho) to test a hyperparameter optimization scheme for phase-picking neural networks and to evaluate their performance. We show that a published model, namely PhaseNet (Zhu and Beroza, 2019), can be simplified and improved with reasonable effort and there are preferred choices of hyperparameters that increase the performance. We also show that models optimized based on the arrival times reported in routine event catalogs consistently perform well when picking arrival times of smaller events, which is crucial for many tasks from microseismicity to explosion monitoring.

58 GEOSCIENCES↗