Search NASA⌕ Search

SEARCH · Search NASA

Results for “Loss function”

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 343 records · Page 19

On the practical usefulness of the Hardware Efficient Ansatz

Variational Quantum Algorithms (VQAs) and Quantum Machine Learning (QML) models train a parametrized quantum circuit to solve a given learning task. The success of these algorithms greatly hinges on appropriately choosing an ansatz for the quantum circuit. Perhaps one of the most famous ansatzes is the one-dimensional layered Hardware Efficient Ansatz (HEA), which seeks to minimize the effect of hardware noise by using native gates and connectives. The use of this HEA has generated a certain ambivalence arising from the fact that while it suffers from barren plateaus at long depths, it can also avoid them at shallow ones. In this work, we attempt to determine whether one should, or should not, use a HEA. We rigorously identify scenarios where shallow HEAs should likely be avoided (e.g., VQA or QML tasks with data satisfying a volume law of entanglement). More importantly, we identify a Goldilocks scenario where shallow HEAs could achieve a quantum speedup: QML tasks with data satisfying an area law of entanglement. We provide examples for such scenario (such as Gaussian diagonal ensemble random Hamiltonian discrimination), and we show that in these cases a shallow HEA is always trainable and that there exists an anti-concentration of loss function values. Our work highlights the crucial role that input states play in the trainability of a parametrized quantum circuit, a phenomenon that is verified in our numerics.

97 MATHEMATICS AND COMPUTING↗

Physics Informed Reinforcement Learning for Power Grid Control using Augmented Random Search

Wide adoption of deep reinforcement learning need to overcome several challenges in energy system domain, including scalability, learning from limited samples, and high-dimensional continuous state and action spaces. In this paper, we integrated physics-based information from the normal generator operation state formula in the reinforcement learning agent's neural network loss function, and applied an augmented random search agent to optimize the generator control under dynamic contingency. Simulation results demonstrated the reliability performance improvements in training speed, reward convergence, sampling efficiency, scalability, and transferability.

physics informed ML, Physics Informed Neural Netwo↗

Bayesian averaging for ground state masses of atomic nuclei in a Machine Learning approach

We present global predictions of the ground state mass of atomic nuclei based on a novel Machine Learning algorithm. We combine precision nuclear experimental measurements together with theoretical predictions of unmeasured nuclei. This hybrid data set is used to train a probabilistic neural network. In addition to training on this data, a physics-based loss function is employed to help refine the solutions. The resultant Bayesian averaged predictions have excellent performance compared to the testing set and come with well-quantified uncertainties which are critical for contemporary scientific applications. We assess extrapolations of the model’s predictions and estimate the growth of uncertainties in the region far from measurements.

74 ATOMIC AND MOLECULAR PHYSICS↗

NSGA-PINN: A Multi-Objective Optimization Method for Physics-Informed Neural Network Training

This paper presents NSGA-PINN, a multi-objective optimization framework for the effective training of physics-informed neural networks (PINNs). The proposed framework uses the non-dominated sorting genetic algorithm (NSGA-II) to enable traditional stochastic gradient optimization algorithms (e.g., ADAM) to escape local minima effectively. Additionally, the NSGA-II algorithm enables satisfying the initial and boundary conditions encoded into the loss function during physics-informed training precisely. We demonstrate the effectiveness of our framework by applying NSGA-PINN to several ordinary and partial differential equation problems. In particular, we show that the proposed framework can handle challenging inverse problems with noisy data.

Lu, Binghang (ORCID:0009000160016632)↗

Short-Term Rainfall Prediction Based on Radar Echo Using an Improved Self-Attention PredRNN Deep Learning Model

Accurate short-term precipitation forecast is extremely important for urban flood warning and natural disaster prevention. In this paper, we present an innovative deep learning model named ISA-PredRNN (improved self-attention PredRNN) for precipitation nowcasting based on radar echoes on the basis of the advanced PredRNN-V2. We introduce the self-attention mechanism and the long-term memory state into the model and design a new set of gating mechanisms. To better capture different intensities of precipitation, the loss function with weights was designed. We further train the model using a combination of reverse scheduled sampling and scheduled sampling to learn the long-term dynamics from the radar echo sequences. Experimental results show that the new model (ISA-PredRNN) can effectively extract the spatiotemporal features of radar echo maps and obtain radar echo prediction results with a small gap from the ground truths. From the comparison with the other six models, the new ISA-PredRNN model has the most accurate prediction results with a critical success index (CSI) of 0.7001, 0.5812 and 0.3052 under the radar echo thresholds of 10 dBZ, 20 dBZ and 30 dBZ, respectively.

Wu, Dali (ORCID:0000000231177074)↗

Physics-Informed Neural Network Solution of Point Kinetics Equations for a Nuclear Reactor Digital Twin

A digital twin (DT) for nuclear reactor monitoring can be implemented using either a differential equations-based physics model or a data-driven machine learning model. The challenge of a physics-model-based DT consists of achieving sufficient model fidelity to represent a complex experimental system, whereas the challenge of a data-driven DT consists of extensive training requirements and a potential lack of predictive ability. We investigate the performance of a hybrid approach, which is based on physics-informed neural networks (PINNs) that encode fundamental physical laws into the loss function of the neural network. We develop a PINN model to solve the point kinetic equations (PKEs), which are time-dependent, stiff, nonlinear, ordinary differential equations that constitute a nuclear reactor reduced-order model under the approximation of ignoring spatial dependence of the neutron flux. The PINN model solution of PKEs is developed to monitor the start-up transient of Purdue University Reactor Number One (PUR-1) using experimental parameters for the reactivity feedback schedule and the neutron source. The results demonstrate strong agreement between the PINN solution and finite difference numerical solution of PKEs. We investigate PINNs performance in both data interpolation and extrapolation. For the test cases considered, the extrapolation errors are comparable to those of interpolation predictions. Extrapolation accuracy decreases with increasing time interval.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Oracle-Preserving Latent Flows

A fundamental task in data science is the discovery, description, and identification of any symmetries present in the data. We developed a deep learning methodology for the simultaneous discovery of multiple non-trivial continuous symmetries across an entire labeled dataset. The symmetry transformations and the corresponding generators are modeled with fully connected neural networks trained with a specially constructed loss function, ensuring the desired symmetry properties. The two new elements in this work are the use of a reduced-dimensionality latent space and the generalization to invariant transformations with respect to high-dimensional oracles. The method is demonstrated with several examples on the MNIST digit dataset, where the oracle is provided by the 10-dimensional vector of logits of a trained classifier. We find classes of symmetries that transform each image from the dataset into new synthetic images while conserving the values of the logits. We illustrate these transformations as lines of equal probability (“flows”) in the reduced latent space. These results show that symmetries in the data can be successfully searched for and identified as interpretable non-trivial transformations in the equivalent latent space.

97 MATHEMATICS AND COMPUTING↗

Active Optical Control with Machine Learning: A Proof of Concept for the Vera C. Rubin Observatory

Abstract The Active Optics System of the Vera C. Rubin Observatory (Rubin) uses information provided by four wave front sensors to determine deviations between the reconstructed wave front and the ideal wave front. The observed deviations are used to adjust the control parameters of the optical system to maintain image quality across the 3.°5 field of view. The baseline approach from the project is to obtain amplitudes of the Zernike polynomials describing the distorted wave front from out-of-focus images collected by the wave front sensors. These Zernike amplitudes are related via an “influence matrix” to the control parameters necessary to correct the wave front. In this paper, we use deep-learning methods to extract the control parameters directly from the images captured by the wave front sensors. Our neural net model uses anti-aliasing pooling to boost performance, and a domain-specific loss function to aid learning and generalization. The accuracy of the control parameters derived from our model exceeds Rubin requirements even in the presence of full-moon background levels and mis-centering of reference stars. Although the training process is time consuming, model evaluation requires only a few milliseconds. This low latency should allow for the correction of the optical configuration during the readout and slew interval between successive exposures.

Astronomy & Astrophysics↗

SpecDis: Value Added Distance Catalog for 4 Million Stars from DESI Year-1 Data

We present the SpecDis value-added stellar distance catalog accompanying DESI Data Release 1. SpecDis trains a feed-forward neural network (NN) with Gaia parallaxes and gets the distance estimates. To build up an unbiased training sample, we do not apply selections on parallax error or signal-to-noise (S/N) of the stellar spectra, and instead, we incorporate parallax error into the loss function. Moreover, we employ principal component analysis to reduce the noise and dimensionality of stellar spectra. Validated by independent external samples of member stars with precise distances from globular clusters, dwarf galaxies, stellar streams, combined with blue horizontal branch stars, we demonstrate that our distance measurements show no significant bias up to 100 kpc, and are much more precise than Gaia parallax beyond 7 kpc. The median distance uncertainties are 23%, 19%, 11%, and 7% for S/N < 20, 20 ≤ S/N < 60, 60 ≤ S/N < 100, and S/N ≥ 100. Selecting stars with ${\mathrm{log}}\,g\lt 3.8$ and distance uncertainties smaller than 25%, we have more than 74,000 giant candidates within 50 kpc of the Galactic center and 1500 candidates beyond this distance. Additionally, we develop a Gaussian mixture model to identify unresolvable equal-mass binaries by modeling the discrepancy between the NN-predicted and the geometric absolute magnitudes from Gaia parallaxes and identify 120,000 equal-mass binary candidates. Our final catalog provides distances and distance uncertainties for >4 million stars, offering a valuable resource for Galactic astronomy.

astronomy data analysis↗

Neural Networks for Nuclear Reactions in MAESTROeX

We demonstrate the use of neural networks to accelerate the reaction steps in the MAESTROeX stellar hydrodynamics code. A traditional MAESTROeX simulation uses a stiff ODE integrator for the reactions; here, we employ a ResNet architecture and describe details relating to the architecture, training, and validation of our networks. Our customized approach includes options for the form of the loss functions, a demonstration that the use of parallel neural networks leads to increased accuracy, and a description of a perturbational approach in the training step that robustifies the model. We test our approach on millimeter-scale flames using a single-step, 3-isotope network describing the first stages of carbon fusion occurring in Type Ia supernovae. We train the neural networks using simulation data from a standard MAESTROeX simulation, and show that the resulting model can be effectively applied to different flame configurations. This work lays the groundwork for more complex networks, and iterative time-integration strategies that can leverage the efficiency of the neural networks.

79 ASTRONOMY AND ASTROPHYSICS↗

On the Convergence of Physics Informed Neural Networks for Linear Second-Order Elliptic and Parabolic Type PDEs

Physics informed neural networks (PINNs) are deep learning based techniques for solving partial differential equations (PDEs) encountered in computational science and engineering. Guided by data and physical laws, PINNs find a neural network that approximates the solution to a system of PDEs. Such a neural network is obtained by minimizing a loss function in which any prior knowledge of PDEs and data are encoded. Despite its remarkable empirical success in one, two or three dimensional problems, there is little theoretical justification for PINNs. As the number of data grows, PINNs generate a sequence of minimizers which correspond to a sequence of neural networks. We want to answer the question: Does the sequence of minimizers converge to the solution to the PDE? We consider two classes of PDEs: linear second-order elliptic and parabolic. By adapting the Schauder approach and the maximum principle, we show that the sequence of minimizers strongly converges to the PDE solution in C 0 . Furthermore, we show that if each minimizer satisfies the initial/boundary conditions, the convergence mode becomes H 1 . Computational examples are provided to illustrate our theoretical findings. To the best of our knowledge, this is the first theoretical work that shows the consistency of PINNs.

97 MATHEMATICS AND COMPUTING↗

Fast and accurate learned multiresolution dynamical downscaling for precipitation

Abstract. This study develops a neural-network-based approach for emulating high-resolution modeled precipitation data with comparable statistical properties but at greatly reduced computational cost. The key idea is to use combination of low- and high-resolution simulations (that differ not only in spatial resolution but also in geospatial patterns) to train a neural network to map from the former to the latter. Specifically, we define two types of CNNs, one that stacks variables directly and one that encodes each variable before stacking, and we train each CNN type both with a conventional loss function, such as mean square error (MSE), and with a conditional generative adversarial network (CGAN), for a total of four CNN variants. We compare the four new CNN-derived high-resolution precipitation results with precipitation generated from original high-resolution simulations, a bilinear interpolater and the state-of-the-art CNN-based super-resolution (SR) technique. Results show that the SR technique produces results similar to those of the bilinear interpolator with smoother spatial and temporal distributions and smaller data variabilities and extremes than the original high-resolution simulations. While the new CNNs trained by MSE generate better results over some regions than the interpolator and SR technique do, their predictions are still biased from the original high-resolution simulations. The CNNs trained by CGAN generate more realistic and physically reasonable results, better capturing not only data variability in time and space but also extremes such as intense and long-lasting storms. The new proposed CNN-based downscaling approach can downscale precipitation from 50 to 12 km in 14 min for 30 years once the network is trained (training takes 4 h using 1 GPU), while the conventional dynamical downscaling would take 1 month using 600 CPU cores to generate simulations at the resolution of 12 km over the contiguous United States.

54 ENVIRONMENTAL SCIENCES↗

DSGAN

This study develops a neural network-based approach for emulating high-resolution modeled precipitation data with comparable statistical properties but at greatly reduced computational cost. The key idea is to use combination of low-and high- resolution simulations to train a neural network to map from the former to the latter. Specifically, we define two types of CNNs, one that stacks variables directly and one that encodes each variable before stacking, and we train each CNN type both with a conventional loss function, such as mean square error (MSE), and with a conditional generative adversarial network (CGAN), for a total of four CNN variants. We compare the four new CNN-derived high-resolution precipitation results with precipitation generated from original high resolution simulations, a bilinear interpolater and the state-of-the-art CNN-based super-resolution (SR) technique. Results show that the SR technique produces results similar to those of the bilinear interpolator with smoother spatial and temporal distributions and smaller data variabilities and extremes than the original high resolution simulations. While the new CNNs trained by MSE generate better results over some regions than the interpolator and SR technique do, their predictions are still not as close as the original high resolution simulations. The CNNs trained by CGAN generate more realistic and physically reasonable results, better capturing not only data variability in time and space but also extremes such as intense and long-lasting storms. The new proposed CNN-based downscaling approach can downscale precipitation from 50~km to 12~km in 14~min for 30~years

LIU, ZHENGCHUN↗

Using support vector machines for anomalous change detonation

We cast anomalous change detection as a binary classification problem, and use a support vector machine (SVM) to build a detector that does not depend on assumptions about the underlying data distribution. To speed up the computation, our SVM is implemented, in part, on a graphical processing unit. Results on real and simulated anomalous changes are used to compare performance to algorithms which effectively assume a Gaussian distribution. In this paper, we investigate the use of support vector machines (SVMs) with radial basis kernels for finding anomalous changes. Compared to typical applications of SVMs, we are operating in a regime of very low false alarm rate. This means that even for relatively large training sets, the data are quite meager in the regime of operational interest. This drives us to use larger training sets, which in turn places more of a computational burden on the SVM. We initially considered three different approaches to to address the need to work in the very low false alarm rate regime. The first is a standard SVM which is trained at one threshold (where more reliable estimates of false alarm rates are possible) and then re-thresholded for the low false alarm rate regime. The second uses the same thresholding approach, but employs a so-called least squares SVM; here a quadratic (instead of a hinge-based) loss function is employed, and for this model, there are good theoretical arguments in favor of adjusting the threshold in a straightforward manner. The third approach employs a weighted support vector machine, where the weights for the two types of errors (false alarm and missed detection) are automatically adjusted to achieve the desired false alarm rate. We have found in previous experiments (not shown here) that the first two types can in some cases work well, while in other cases they do not. This renders both approaches unreliable for automated change detection. By contrast, the third approach reliably produces good results, but at the cost of larger computational requirements caused by the need to estimate very small false alarm rates. To address these computational requirements, we employ a recently developed in-house solver for SVMs that is significantly faster than freely available standard solvers. But these computational issues are secondary to the larger question: do kernelized solutions provide better performance, in terms of detection rates and false alarm rates, than more traditional methods for change detection that effectively assume Gaussian data distributions? To this end, we will compare ROC curves obtained from the SVM with those from chronochrome, covariance equalization, and hyperbolic anomalous change detection.

97 MATHEMATICS AND COMPUTING↗

Novel deep learning methods for track reconstruction

For the past year, the HEP.TrkX project has been investigating machine learning solutions to LHC particle track reconstruction problems. A variety of models were studied that drew inspiration from computer vision applications and operated on an image-like representation of tracking detector data. While these approaches have shown some promise, image-based methods face challenges in scaling up to realistic HL-LHC data due to high dimensionality and sparsity. In contrast, models that can operate on the spacepoint representation of track measurements ("hits") can exploit the structure of the data to solve tasks efficiently. In this paper we will show two sets of new deep learning models for reconstructing tracks using space-point data arranged as sequences or connected graphs. In the first set of models, Recurrent Neural Networks (RNNs) are used to extrapolate, build, and evaluate track candidates akin to Kalman Filter algorithms. Such models can express their own uncertainty when trained with an appropriate likelihood loss function. The second set of models use Graph Neural Networks (GNNs) for the tasks of hit classification and segment classification. These models read a graph of connected hits and compute features on the nodes and edges. They adaptively learn which hit connections are important and which are spurious. The models are scaleable with simple architecture and relatively few parameters. Results for all models will be presented on ACTS generic detector simulated data.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

A Physics-Driven Deep-Learning Network for Solving Nonlinear Inverse Problems

Geosteering inversion, which can be viewed as a nonlinear inverse problem, is an important technique used by directional drilling. Traditional methods that rely on iterative procedures and regularization are sensitive to the selection of initial values and can be slow due to convergence issues. In industrial applications, a lookup table is used to produce fast predictions. However, performance is not guaranteed by this approach due to the limitation of the hardware. In this paper, we propose a novel physics-driven deep-learning framework for providing a fast, accurate surrogate to solve the inverse problem. Particularly, leveraged by the forward physical model and 1D convolutional neural network (1D-CNN), the proposed method provides more reliable solutions to the inverse problem with improved performance. In addition, a new physics-driven loss function is introduced to accommodate both the model misfit and the data misfit. In conclusion, our experiments demonstrate the effectiveness of our method.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Anderson Acceleration for Distributed Training of Deep Learning Models

Anderson acceleration (AA) is an extrapolation technique that has recently gained interest in the deep learning (DL) community to speed-up the sequential training of DL models. However, when performed at large scale, the DL training is exposed to a higher risk of getting trapped into steep local minima of the training loss function, and standard AA does not provide sufficient acceleration to escape from these steep local minima. This results in poor generalizability and makes AA ineffective. To restore AA’s advantage to speed-up the training of DL models on large scale computing platforms, we combine AA with an adaptive moving average procedure that boosts the training to escape from steep local minima. By monitoring the relative standard deviation between consecutive iterations, we also introduce a criterion to automatically assess whether the moving average is needed. We applied the method to the following DL instantiations for image classification: (i) ResNet50 trained on the open-source CIFAR100 dataset and (ii) ResNet50 trained on the open-source ImageNet1k dataset. Numerical results obtained using up to 1,536 NVIDIA V100 GPUs on the OLCF supercomputer Summit showed the stabilizing effect of the moving average on AA for all the problems above.

Lupo Pasini, Massimiliano↗

Refining fast simulation using machine learning

At the CMS experiment, a growing reliance on the fast Monte Carlo application (FastSim) will accompany the high luminosity and detector granularity expected in Phase 2. The FastSim chain is roughly 10 times faster than the application based on the GEANT4 detector simulation and full reconstruction referred to as FullSim. However, this advantage comes at the price of decreased accuracy in some of the final analysis observables. In this contribution, a machine learning-based technique to refine those observables is presented. We employ a regression neural network trained with a sophisticated combination of multiple loss functions to provide post-hoc corrections to samples produced by the FastSim chain. The results show considerably improved agreement with the FullSim output and an improvement in correlations among output observables and external parameters. This technique is a promising replacement for existing correction factors, providing higher accuracy and thus contributing to the wider usage of FastSim.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗