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 937 records · Page 52

Devices and methods for increasing the speed or power efficiency of a computer when performing machine learning using spiking neural networks

A method for increasing a speed and efficiency of a computer when performing machine learning using spiking neural networks. The method includes computer-implemented operations; that is, operations that are solely executed on a computer. The method includes receiving, in a spiking neural network, a plurality of input values upon which a machine learning algorithm is based. The method also includes correlating, for each input value, a corresponding response speed of a corresponding neuron to a corresponding equivalence relationship between the input value to a corresponding latency of the corresponding neuron. Neurons that trigger faster than other neurons represent close relationships between input values and neuron latencies. Latencies of the neurons represent data points used in performing the machine learning. A plurality of equivalence relationships are formed as a result of correlating. The method includes performing the machine learning using the plurality of equivalence relationships.

Vineyard, Craig Michael↗

Inversion of parameters for semiarid regions by a neural network

Microwave brightness temperatures obtained from a passive radiative transfer model are inverted through use of a neural network. The model is applicable to semiarid regions and produces dual-polarized brightness temperatures for 6.6-, 10.7-, and 37-GHz frequencies. A range of temperatures is generated by varying three geophysical parameters over acceptable ranges: soil moisture, vegetation moisture, and soil temperature. A multilayered perceptron (MLP) neural network is trained with a subset of the generated temperatures, and the remaining temperatures are inverted using a backpropagation method. Several synthetic terrains are devised and inverted by the network under local constraints. All the inversions show good agreement with the original geophysical parameters, falling within 5 percent of the actual value of the parameter range.

Zurk, Lisa M.↗

Learning characteristics of a space-time neural network as a tether skiprope observer

The Software Technology Laboratory at JSC is testing a Space Time Neural Network (STNN) for observing tether oscillations present during retrieval of a tethered satellite. Proper identification of tether oscillations, known as 'skiprope' motion, is vital to safe retrieval of the tethered satellite. Our studies indicate that STNN has certain learning characteristics that must be understood properly to utilize this type of neural network for the tethered satellite problem. We present our findings on the learning characteristics including a learning rate versus momentum performance table.

Lea, Robert N.↗

Using neural networks and Dyna algorithm for integrated planning, reacting and learning in systems

The traditional AI answer to the decision making problem for a robot is planning. However, planning is usually CPU-time consuming, depending on the availability and accuracy of a world model. The Dyna system generally described in earlier work, uses trial and error to learn a world model which is simultaneously used to plan reactions resulting in optimal action sequences. It is an attempt to integrate planning, reactive, and learning systems. The architecture of Dyna is presented. The different blocks are described. There are three main components of the system. The first is the world model used by the robot for internal world representation. The input of the world model is the current state and the action taken in the current state. The output is the corresponding reward and resulting state. The second module in the system is the policy. The policy observes the current state and outputs the action to be executed by the robot. At the beginning of program execution, the policy is stochastic and through learning progressively becomes deterministic. The policy decides upon an action according to the output of an evaluation function, which is the third module of the system. The evaluation function takes the following as input: the current state of the system, the action taken in that state, the resulting state, and a reward generated by the world which is proportional to the current distance from the goal state. Originally, the work proposed was as follows: (1) to implement a simple 2-D world where a 'robot' is navigating around obstacles, to learn the path to a goal, by using lookup tables; (2) to substitute the world model and Q estimate function Q by neural networks; and (3) to apply the algorithm to a more complex world where the use of a neural network would be fully justified. In this paper, the system design and achieved results will be described. First we implement the world model with a neural network and leave Q implemented as a look up table. Next, we use a lookup table for the world model and implement the Q function with a neural net. Time limitations prevented the combination of these two approaches. The final section discusses the results and gives clues for future work.

Lima, Pedro↗

Self-adaptive weights based on balanced residual decay rate for physics-informed neural networks and deep operator networks

Physics-informed deep learning has emerged as a promising alternative for solving partial differential equations. However, for complex problems, training these networks can still be challenging, often resulting in unsatisfactory accuracy and efficiency. In this work, we demonstrate that the failure of plain physics-informed neural networks arises from the significant discrepancy in the convergence rate of residuals at different training points, where the slowest convergence rate dominates the overall solution convergence. Based on these observations, we propose a pointwise adaptive weighting method that balances the residual decay rate across different training points. The performance of our proposed adaptive weighting method is compared with current state-of-the-art adaptive weighting methods on benchmark problems for both physics-informed neural networks and physics-informed deep operator networks. In conclusion, through extensive numerical results we demonstrate that our proposed approach of balanced residual decay rates offers several advantages, including bounded weights, high prediction accuracy, fast convergence rate, low training uncertainty, low computational cost, and ease of hyperparameter tuning.

Balanced convergence rate↗

Entropy Stable Conservative Flux Form Neural Networks

We propose an entropy-stable conservative flux form neural network (CFN) to predict the dynamics of unknown governing conservation laws. The design of the network is based on the entropy-stable, second-order, and non-oscillatory Kurganov-Tadmor (KT) scheme. The proposed entropy-stable CFN, hereafter referred to as ESCFN, uses slope limiting as a denoising mechanism, ensuring accurate predictions in both noisy and sparse observation environments, as well as in both smooth and discontinuous regions. Importantly, our method is designed to predict long term dynamics of the unknown conservation law exclusively from a short temporal window of observed data, that is, without oracle knowledge of the PDE or later-time solution profiles. Numerical experiments demonstrate that the ESCFN achieves both stability and conservation while maintaining accuracy over extended time domains, and successfully predicts shock propagation speeds in long-term simulations. Furthermore, it is also robust to both noisy and sparse data environments.

Hyperbolic conservation laws↗

Convolutional Neural Networks for Problems in Transport Phenomena: A Theoretical Minimum

Convolutional neural network (CNN), a deep learning algorithm, has gained popularity in technological applications that rely on interpreting images (typically, an image is a 2D field of pixels). Transport phenomena is the science of studying different fields representing mass, momentum, or heat transfer. Some of the common fields are species concentration, fluid velocity, pressure, and temperature. Each of these fields can be expressed as an image(s). Consequently, CNNs can be leveraged to solve specific scientific problems in transport phenomena. Herein, we show that such problems can be grouped into three basic categories: (a) mapping a field to a descriptor (b) mapping a field to another field, and (c) mapping a descriptor to a field. After reviewing the representative transport phenomena literature for each of these categories, we illustrate the necessary steps for constructing appropriate CNN solutions using sessile liquid drops as an exemplar problem. If sufficient training data is available, CNNs can considerably speed up the solution of the corresponding problems. Finally, the present discussion is meant to be minimalistic such that readers can easily identify the transport phenomena problems where CNNs can be useful as well as construct and/or assess such solutions.

97 MATHEMATICS AND COMPUTING↗

Neural network uncertainty assessment using Bayesian statistics: a remote sensing application

Neural network (NN) techniques have proved successful for many regression problems, in particular for remote sensing; however, uncertainty estimates are rarely provided. In this article, a Bayesian technique to evaluate uncertainties of the NN parameters (i.e., synaptic weights) is first presented. In contrast to more traditional approaches based on point estimation of the NN weights, we assess uncertainties on such estimates to monitor the robustness of the NN model. These theoretical developments are illustrated by applying them to the problem of retrieving surface skin temperature, microwave surface emissivities, and integrated water vapor content from a combined analysis of satellite microwave and infrared observations over land. The weight uncertainty estimates are then used to compute analytically the uncertainties in the network outputs (i.e., error bars and correlation structure of these errors). Such quantities are very important for evaluating any application of an NN model. The uncertainties on the NN Jacobians are then considered in the third part of this article. Used for regression fitting, NN models can be used effectively to represent highly nonlinear, multivariate functions. In this situation, most emphasis is put on estimating the output errors, but almost no attention has been given to errors associated with the internal structure of the regression model. The complex structure of dependency inside the NN is the essence of the model, and assessing its quality, coherency, and physical character makes all the difference between a blackbox model with small output errors and a reliable, robust, and physically coherent model. Such dependency structures are described to the first order by the NN Jacobians: they indicate the sensitivity of one output with respect to the inputs of the model for given input data. We use a Monte Carlo integration procedure to estimate the robustness of the NN Jacobians. A regularization strategy based on principal component analysis is proposed to suppress the multicollinearities in order to make these Jacobians robust and physically meaningful.

Neural Networks (Computer)↗

Enhancing high-fidelity neural network potentials through low-fidelity sampling

The efficacy of neural network potentials (NNPs) critically depends on the quality of the configurational datasets used for training. Prior research using empirical potentials has shown that well-selected liquid–solid transitional configurations of a metallic system can be translated to other metallic systems. This study demonstrates that such validated configurations can be relabeled using density functional theory (DFT) calculations, thereby enhancing the development of high-fidelity NNPs. Training strategies and sampling approaches are efficiently assessed using empirical potentials and subsequently relabeled via DFT in a highly parallelized fashion for high-fidelity NNP training. Our results reveal that relying solely on energy and force for NNP training is inadequate to prevent overfitting, highlighting the necessity of incorporating stress terms into the loss functions. To optimize training involving force and stress terms, we propose employing transfer learning to fine-tune the weights, ensuring that the potential surface is smooth for these quantities composed of energy derivatives. This approach markedly improves the accuracy of elastic constants derived from simulations in both empirical potential-based NNPs and relabeled DFT-based NNPs. Overall, this study offers significant insights into leveraging empirical potentials to expedite the development of reliable and robust NNPs at the DFT level.

97 MATHEMATICS AND COMPUTING↗

Forecasting of preprocessed daily solar radiation time series using neural networks

In this paper, we present an application of Artificial Neural Networks (ANNs) in the renewable energy domain. We particularly look at the Multi-Layer Perceptron (MLP) network which has been the most used of ANNs architectures both in the renewable energy domain and in the time series forecasting. We have used a MLP and an ad hoc time series pre-processing to develop a methodology for the daily prediction of global solar radiation on a horizontal surface. First results are promising with nRMSE {proportional_to} 21% and RMSE {proportional_to} 3.59 MJ/m{sup 2}. The optimized MLP presents predictions similar to or even better than conventional and reference methods such as ARIMA techniques, Bayesian inference, Markov chains and k-Nearest-Neighbors. Moreover we found that the data pre-processing approach proposed can reduce significantly forecasting errors of about 6% compared to conventional prediction methods such as Markov chains or Bayesian inference. The simulator proposed has been obtained using 19 years of available data from the meteorological station of Ajaccio (Corsica Island, France, 41 55'N, 8 44'E, 4 m above mean sea level). The predicted whole methodology has been validated on a 1.175 kWc mono-Si PV power grid. Six prediction methods (ANN, clear sky model, combination..) allow to predict the best daily DC PV power production at horizon d + 1. The cumulated DC PV energy on a 6-months period shows a great agreement between simulated and measured data (R{sup 2} > 0.99 and nRMSE < 2%). (author)

14 SOLAR ENERGY↗

SwitchX : Gmin-Gmax Switching for Energy-efficient and Robust Implementation of Binarized Neural Networks on ReRAM Xbars

Memristive crossbars can efficiently implement Binarized Neural Networks (BNNs) wherein the weights are stored in high-resistance states (HRS) and low-resistance states (LRS) of the synapses. We propose SwitchX mapping of BNN weights onto ReRAM crossbars such that the impact of crossbar non-idealities, that lead to degradation in computational accuracy, are minimized. Essentially, SwitchX maps the binary weights in such a manner that a crossbar instance comprises of more HRS than LRS synapses. We find BNNs mapped onto crossbars with SwitchX to exhibit better robustness against adversarial attacks than the standard crossbar mapped BNNs, the baseline. Finally, we combine SwitchX with state-aware training (that further increases the feasibility of HRS states during weight mapping) to boost the robustness of a BNN on hardware. We find that this approach yields stronger defense against adversarial attacks than adversarial training, a state-of the-art software defense. We perform experiments on a VGG16 BNN with benchmark datasets (CIFAR-10, CIFAR-100 and TinyImagenet) and use Fast Gradient Sign Method (ϵ = 0.05 to 0.3) and Projected Gradient Descent (ϵ = $\frac{2}{255}$ to $\frac{32}{255}$, α = $\frac{2}{255}$) adversarial attacks. We show that SwitchX combined with state-aware training can yield upto ~35% improvements in clean accuracy and ~6–16% in adversarial accuracies against conventional BNNs. Furthermore, an important by-product of SwitchX mapping is increased crossbar power savings, owing to an increased proportion of HRS synapses, which is furthered with state-aware training. We obtain upto ~21–22% savings in crossbar power consumption for state-aware trained BNN mapped via SwitchX on 16 × 16 and 32 × 32 crossbars using the CIFAR-10 and CIFAR-100 datasets.

97 MATHEMATICS AND COMPUTING↗

Full-dimensional and reduced-dimensional calculations of initial state-selected reaction probabilities studying the H + CH{sub 4} → H{sub 2} + CH{sub 3} reaction on a neural network PES

Initial state-selected reaction probabilities of the H + CH{sub 4} → H{sub 2} + CH{sub 3} reaction are calculated in full and reduced dimensionality on a recent neural network potential [X. Xu, J. Chen, and D. H. Zhang, Chin. J. Chem. Phys. 27, 373 (2014)]. The quantum dynamics calculation employs the quantum transition state concept and the multi-layer multi-configurational time-dependent Hartree approach and rigorously studies the reaction for vanishing total angular momentum (J = 0). The calculations investigate the accuracy of the neutral network potential and study the effect resulting from a reduced-dimensional treatment. Very good agreement is found between the present results obtained on the neural network potential and previous results obtained on a Shepard interpolated potential energy surface. The reduced-dimensional calculations only consider motion in eight degrees of freedom and retain the C{sub 3v} symmetry of the methyl fragment. Considering reaction starting from the vibrational ground state of methane, the reaction probabilities calculated in reduced dimensionality are moderately shifted in energy compared to the full-dimensional ones but otherwise agree rather well. Similar agreement is also found if reaction probabilities averaged over similar types of vibrational excitation of the methane reactant are considered. In contrast, significant differences between reduced and full-dimensional results are found for reaction probabilities starting specifically from symmetric stretching, asymmetric (f{sub 2}-symmetric) stretching, or e-symmetric bending excited states of methane.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Accelerated Charged Particle Tracking with Graph Neural Networks on FPGAs

We develop and study FPGA implementations of algorithms for charged particle tracking based on graph neural networks. The two complementary FPGA designs are based on OpenCL, a framework for writing programs that execute across heterogeneous platforms, and hls4ml, a high-level-synthesis-based compiler for neural network to firmware conversion. We evaluate and compare the resource usage, latency, and tracking performance of our implementations based on a benchmark dataset. We find a considerable speedup over CPU-based execution is possible, potentially enabling such algorithms to be used effectively in future computing workflows and the FPGA-based Level-1 trigger at the CERN Large Hadron Collider.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

AutoTandemML: Active Learning Enhanced Tandem Neural Networks for Inverse Design Problems

Inverse design in science and engineering involves determining optimal design parameters that achieve desired performance outcomes, a process often hindered by the complexity and high dimensionality of design spaces, leading to significant computational costs. To tackle this challenge, we propose a novel hybrid approach that combines active learning with Tandem Neural Networks to enhance the efficiency and effectiveness of solving inverse design problems. Active learning allows to selectively sample the most informative data points, reducing the required dataset size without compromising accuracy. We investigate this approach using three benchmark problems: airfoil inverse design, photonic surface inverse design, and scalar boundary condition reconstruction in diffusion partial differential equations. We demonstrate that integrating active learning with Tandem Neural Networks outperforms standard approaches across the benchmark suite, achieving better accuracy with fewer training samples.

97 MATHEMATICS AND COMPUTING↗

Fast inference of deep neural networks in FPGAs for particle physics

Here, recent results at the Large Hadron Collider (LHC) have pointed to enhanced physics capabilities through the improvement of the real-time event processing techniques. Machine learning methods are ubiquitous and have proven to be very powerful in LHC physics, and particle physics as a whole. However, exploration of the use of such techniques in low-latency, low-power FPGA (Field Programmable Gate Array) hardware has only just begun. FPGA-based trigger and data acquisition systems have extremely low, sub-microsecond latency requirements that are unique to particle physics. We present a case study for neural network inference in FPGAs focusing on a classifier for jet substructure which would enable, among many other physics scenarios, searches for new dark sector particles and novel measurements of the Higgs boson. While we focus on a specific example, the lessons are far-reaching. A companion compiler package for this work is developed based on High-Level Synthesis (HLS) called hls4ml to build machine learning models in FPGAs. The use of HLS increases accessibility across a broad user community and allows for a drastic decrease in firmware development time. We map out FPGA resource usage and latency versus neural network hyperparameters to identify the problems in particle physics that would benefit from performing neural network inference with FPGAs. For our example jet substructure model, we fit well within the available resources of modern FPGAs with a latency on the scale of 100 ns.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Effective cosmic density field reconstruction with convolutional neural network

ABSTRACT We present a cosmic density field reconstruction method that augments the traditional reconstruction algorithms with a convolutional neural network (CNN). Following previous work, the key component of our method is to use the reconstructed density field as the input to the neural network. We extend this previous work by exploring how the performance of these reconstruction ideas depends on the input reconstruction algorithm, the reconstruction parameters, and the shot noise of the density field, as well as the robustness of the method. We build an eight-layer CNN and train the network with reconstructed density fields computed from the Quijote suite of simulations. The reconstructed density fields are generated by both the standard algorithm and a new iterative algorithm. In real space at z = 0, we find that the reconstructed field is 90 per cent correlated with the true initial density out to $k\sim 0.5 \, \mathrm{ h}\, \rm {Mpc}^{-1}$, a significant improvement over $k\sim 0.2 \, \mathrm{ h}\, \rm {Mpc}^{-1}$ achieved by the input reconstruction algorithms. We find similar improvements in redshift space, including an improved removal of redshift space distortions at small scales. We also find that the method is robust across changes in cosmology. Additionally, the CNN removes much of the variance from the choice of different reconstruction algorithms and reconstruction parameters. However, the effectiveness decreases with increasing shot noise, suggesting that such an approach is best suited to high density samples. This work highlights the additional information in the density field beyond linear scales as well as the power of complementing traditional analysis approaches with machine learning techniques.

Astronomy & Astrophysics↗

Electronic neural network for dynamic resource allocation

A VLSI implementable neural network architecture for dynamic assignment is presented. The resource allocation problems involve assigning members of one set (e.g. resources) to those of another (e.g. consumers) such that the global 'cost' of the associations is minimized. The network consists of a matrix of sigmoidal processing elements (neurons), where the rows of the matrix represent resources and columns represent consumers. Unlike previous neural implementations, however, association costs are applied directly to the neurons, reducing connectivity of the network to VLSI-compatible 0 (number of neurons). Each row (and column) has an additional neuron associated with it to independently oversee activations of all the neurons in each row (and each column), providing a programmable 'k-winner-take-all' function. This function simultaneously enforces blocking (excitatory/inhibitory) constraints during convergence to control the number of active elements in each row and column within desired boundary conditions. Simulations show that the network, when implemented in fully parallel VLSI hardware, offers optimal (or near-optimal) solutions within only a fraction of a millisecond, for problems up to 128 resources and 128 consumers, orders of magnitude faster than conventional computing or heuristic search methods.

Thakoor, A. P.↗