Search NASA⌕ Search

SEARCH · Search NASA

Results for “Generalization error”

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 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators

It is widely known that neural networks (NNs) are universal approximators of continuous functions. However, a less known but powerful result is that a NN with a single hidden layer can accurately approximate any nonlinear continuous operator. This universal approximation theorem of operators is suggestive of the structure and potential of deep neural networks (DNNs) in learning continuous operators or complex systems from streams of scattered data. Here, in this work, we thus extend this theorem to DNNs. We design a new network with small generalization error, the deep operator network (DeepONet), which consists of a DNN for encoding the discrete input function space (branch net) and another DNN for encoding the domain of the output functions (trunk net). We demonstrate that DeepONet can learn various explicit operators, such as integrals and fractional Laplacians, as well as implicit operators that represent deterministic and stochastic differential equations. We study different formulations of the input function space and its effect on the generalization error for 16 different diverse applications.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2019

Learning concave-convex profiles of data transport over dedicated connections

Dedicated data transport infrastructures are increasingly being deployed to support distributed big-data and high-performance computing scenarios. These infrastructures employ data transfer nodes that use sophisticated software stacks to support network transport among sites, which often house distributed file and storage systems. Throughput measurements collected over such infrastructures for a range of round trip times (RTTs) reflect the underlying complex end-to-end connections, and have revealed dichotomous throughput profiles as functions of RTT. In particular, concave regions of throughput profiles at lower RTTs indicate near-optimal performance, and convex regions at higher RTTs indicate bottlenecks due to factors such as buffer or credit limits. We present a machine learning method that explicitly infers these concave and convex regions and transitions between them using sigmoid functions. We also provide distribution-free confidence estimates for the generalization error of these concave-convex profile estimates. Throughput profiles for data transfers over 10 Gbps connections with 0–366 ms RTT provide important performance insights, including the near optimality of transfers performed with the XDD tool between XFS filesystems, and the performance limits of wide-area Lustre extensions using LNet routers. A direct application of generic machine learning packages does not adequately highlight these critical performance regions or provide as precise confidence estimates.

Rao, Nageswara↗
DOE OSTI2025

Scalable Hybrid Learning Techniques for Scientific Data Compression

Data compression is becoming critical for storing scientific data because many scientific applications need to store large amounts of data and post process this data for scientific discovery. Unlike image and video compression algorithms that limit errors to primary data (PD), scientists require compression techniques that accurately preserve derived quantities of interest (QoIs). Here, this article presents a physics-informed compression technique implemented as an end-to-end, scalable, GPU-based pipeline for data compression that addresses this requirement. Our hybrid compression technique combines machine learning techniques and standard compression methods. Specifically, we combine an autoencoder, an error-bounded lossy compressor to provide guarantees on raw data error, and a constraint satisfaction post-processing step to preserve the QoIs within a minimal error (generally less than floating point error). The effectiveness of the data compression pipeline is demonstrated by compressing nuclear fusion simulation data generated by a large-scale fusion code, XGC, which produces hundreds of terabytes of data in a single day. Our approach works within the ADIOS framework and results in compression by a factor of more than 150 while requiring only a few percent of the computational resources necessary for generating the data, making the overall approach highly effective for practical scenarios.

ITER↗
NASA NTRS2006

TRMM On-Orbit Performance Re-Accessed After Control Change

The Tropical Rainfall Measuring Mission (TRMM) spacecraft, a joint mission between the U.S. and Japan, launched onboard an HI1 rocket on November 27,1997 and transitioned in August, 2001 from an average operating altitude of 350 kilometers to 402.5 kilometers. Due to problems using the Earth Sensor Assembly (ESA) at the higher altitude, TRMM switched to a backup attitude control mode. Prior to the orbit boost TRMM controlled pitch and roll to the local vertical using ESA measurements while using gyro data to propagate yaw attitude between yaw updates from the Sun sensors. After the orbit boost, a Kalman filter used 3-axis gyro data with Sun sensor and magnetometers to estimate onboard attitude. While originally intended to meet a degraded attitude accuracy of 0.7 degrees, the new control mode met the original 0.2 degree attitude accuracy requirement after improving onboard ephemeris prediction and adjusting the magnetometer calibration onboard. Independent roll attitude checks using a science instrument, the Precipitation Radar (PR) which was built in Japan, provided a novel insight into the pointing performance. The PR data helped identify the pointing errors after the orbit boost, track the performance improvements, and show subtle effects from ephemeris errors and gyro bias errors. It also helped identify average bias trends throughout the mission. Roll errors tracked by the PR from sample orbits pre-boost and post-boost are shown in Figure 1. Prior to the orbit boost the largest attitude errors were due to occasional interference in the ESA. These errors were sometime larger than 0.2 degrees in pitch and roll, but usually less, as estimated from a comprehensive review of the attitude excursions using gyro data. Sudden jumps in the onboard roll show up as spikes in the reported attitude since the control responds within tens of seconds to null the pointing error. The PR estimated roll tracks well with an estimate of the roll history propagated using gyro data. After the orbit boost, the attitude errors shown by the PR roll have a smooth sine-wave type signal because of the way that attitude errors propagate with the use of gyro data. Yaw errors couple at orbit period to roll with '/4 orbit lag. By tracking the amplitude, phase, and bias of the sinusoidal PR roll error signal, it was shown that the average pitch rotation axis tends to be offset from orbit normal in a direction perpendicular to the Sun direction, as shown in Figure 2 for a 200 day period following the orbit boost. This is a result of the higher accuracy and stability of the Sun sensor measurements relative to the magnetometer measurements used in the Kalman filter. In November, 2001 a magnetometer calibration adjustment was uploaded which improved the pointing performance, keeping the roll and yaw amplitudes within about 0.1 degrees. After the boost, onboard ephemeris errors had a direct effect on the pitch pointing, being used to compute the Earth pointing reference frame. Improvements after the orbit boost have kept the the onboard ephemeris errors generally below 20 kilometers. Ephemeris errors have secondary effects on roll and yaw, especially during high beta angle when pitch effects can couple into roll and yaw. This is illustrated in figure 3. The onboard roll bias trends as measured by PR data show correlations with the Kalman filter's gyro bias error. This particularly shows up after yaw turns (every 2 to 4 weeks) as shown in Figure 3, when a slight roll bias is observed while the onboard computed gyro biases settle to new values. As for longer term trends, the PR data shows that the roll bias was influenced by Earth horizon radiance effects prior to the boost, changing values at yaw turns, and indicated a long term drift as shown in Figure 4. After the boost, the bias variations were smaller and showed some possible correlation with solar beta angle, probably due to sun sensor misalignment effects.

Bilanow, Steve↗
NASA NTRS1973

Error analysis of multi-conic techniques

A general error analysis of three recently developed multi-conic methods of three-body trajectory integration has been carried out. Single-step error functions for position and velocity have been derived as Taylor series in powers of the time step and also in integral form. These error functions are used to investigate the relative accuracy of the three methods in various regions of the earth-moon space and to provide a method of variable step size control for the trajectory integration procedure. Numerical results are used to compare the multi-step performance of the methods for both large and small step sizes.

D'Amario, L. A.↗
NASA NTRS1981

A general analysis of anti-jam communication systems

A general error bound is derived for a general anti-jam communication system which will serve as the basis for evaluating the performance of all such complex communication systems. The two most common spread spectrum techniques, coherent DS/BPSK and noncoherent FH/MFSK, are analyzed. Pulse jamming represents the worst type of jammer for DS/BPSK systems, and several receiver structures against such a jammer are examined. It is found that for low values of chip energy-to-noise ratios of O dB or less there is little difference between having or not having jammer state knowledge with a hard decision receiver. Soft decision receivers are shown to be useless against very narrow pulses without jammer state knowledge. Partial band jammers are close to the worst case jammer for FH/MFSK systems. The conclusions found for these systems are similar to those for the DS/BPSK systems.

Omura, J. K.↗
DOE OSTI2022

On Feature, Classifier and Detector Fusers for 235U Signatures Using Gamma Spectral Counts

Three types of information fusion strategies are studied to assess the performance of classifiers for detecting low-level 235 U radiation sources, using features obtained from gamma spectra of NaI detectors. These three strategies are based on using two spectral region features, fusing eight classifiers of diverse designs, and fusing multiple detectors located at different positions around the source. The inner, middle and outer groups of detectors, within a formation of two concentric circles and a spiral of 21 detectors, are identified based on their distance to the source, which is located at the center. This study provides two main qualitative insights into this classification task. First, the fusion of detectors leads to an overall improved classification performance, least in the inner group, most in the outer group, and in between for the middle group. Second, several classifiers and fusers achieve lower training error which does not translate to lower generalization error, indicating their over-fitting to training data.

Rao, Nageswara↗
DOE OSTI2022

On Feature, Classifier and Detector Fusers for U-235 Signatures Using Gamma Spectral Counts

Three types of information fusion strategies are studied to assess the performance of classifiers for detecting low-level U-235 radiation sources, using features obtained from gamma spectra of NaI detectors. These three strategies are based on using two spectral region features, fusing eight classifiers of diverse designs, and fusing multiple detectors located at different positions around the source. The inner, middle and outer groups of detectors, within a formation of two concentric circles and a spiral of 21 detectors, are identified based on their distance to the source, which is located at the center. This study provides two main qualitative insights into this classification task. First,the fusion of detectors leads to an overall improved classification performance, least in the inner group, most in the outer group, and in between for the middle group. Second, several classifiers and fusers achieve lower training error which does not translate to lower generalization error, indicating their over-fitting to training data.

Rao, Nageswara↗
DOE OSTI2024

On the Training and Generalization of Deep Operator Networks

Here, we present a novel training method for deep operator networks (DeepONets), one of the most popular neural network models for operators. DeepONets are constructed by two subnetworks, namely the branch and trunk networks. Typically, the two subnetworks are trained simultaneously, which amounts to solving a complex optimization problem in a high dimensional space. In addition, the nonconvex and nonlinear nature makes training very challenging. To tackle such a challenge, we propose a two-step training method that trains the trunk network first and then sequentially trains the branch network. The core mechanism is motivated by the divide-and-conquer paradigm and is the decomposition of the entire complex training task into two subtasks with reduced complexity. Therein the Gram–Schmidt orthonormalization process is introduced which significantly improves stability and generalization ability. On the theoretical side, we establish a generalization error estimate in terms of the number of training data, the width of DeepONets, and the number of input and output sensors. Numerical examples are presented to demonstrate the effectiveness of the two-step training method, including Darcy flow in heterogeneous porous media.

deep operator networks↗
DOE OSTI2025

Coefficient-to-Basis Network: a fine-tunable operator learning framework for inverse problems with adaptive discretizations and theoretical guarantees

We propose a Coefficient-to-Basis Network (C2BNet), a novel framework for solving inverse problems within the operator learning paradigm. C2BNet efficiently adapts to different discretizations through fine-tuning, using a pre-trained model to significantly reduce computational cost while maintaining high accuracy. Unlike traditional approaches that require retraining from scratch for new discretizations, our method enables seamless adaptation without sacrificing predictive performance. Furthermore, we establish theoretical approximation and generalization error bounds for C2BNet by exploiting low-dimensional structures in the underlying datasets. Our analysis demonstrates that C2BNet adapts to low-dimensional structures without relying on explicit encoding mechanisms, highlighting its robustness and efficiency. To validate our theoretical findings, we conducted extensive numerical experiments that showcase the superior performance of C2BNet on several inverse problems. The results confirm that C2BNet effectively balances computational efficiency and accuracy, making it a promising tool to solve inverse problems in scientific computing and engineering applications.

97 MATHEMATICS AND COMPUTING↗
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↗
DOE OSTI2020

NSFnets (Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations

In the last 50 years there has been a tremendous progress in solving numerically the Navier-Stokes equations using finite differences, finite elements, spectral, and even meshless methods. Yet, in many real cases, we still cannot incorporate seamlessly (multi-fidelity) data into existing algorithms, and for industrial-complexity applications the mesh generation is time consuming and still an art. Moreover, solving ill-posed problems (e.g., lacking boundary conditions) or inverse problems is often prohibitively expensive and requires different formulations and new computer codes. Here, we employ physics-informed neural networks (PINNs), encoding the governing equations directly into the deep neural network via automatic differentiation, to overcome some of the aforementioned limitations for simulating incompressible laminar and turbulent flows. We develop the Navier-Stokes flow nets (NSFnets) by considering two different mathematical formulations of the Navier-Stokes equations: the velocity-pressure (VP) formulation and the vorticity-velocity (VV) formulation. Since this is a new approach, we first select some standard benchmark problems to assess the accuracy, convergence rate, computational cost and flexibility of NSFnets; analytical solutions and direct numerical simulation (DNS) databases provide proper initial and boundary conditions for the NSFnet simulations. The spatial and temporal coordinates are the inputs of the NSFnets, while the instantaneous velocity and pressure fields are the outputs for the VP-NSFnet, and the instantaneous velocity and vorticity fields are the outputs for the VV-NSFnet. This is unsupervised learning and, hence, no labeled data are required beyond boundary and initial conditions and the fluid properties. The residuals of the VP or VV governing equations, together with the initial and boundary conditions, are embedded into the loss function of the NSFnets. No data is provided for the pressure to the VP-NSFnet, which is a hidden state and is obtained via the incompressibility constraint without extra computational cost. Unlike the traditional numerical methods, NSFnets inherit the properties of neural networks (NNs), hence the total error is composed of the approximation, the optimization, and the generalization errors. Here, we empirically attempt to quantify these errors by varying the sampling (“residual”) points, the iterative solvers, and the size of the NN architecture. For the laminar flow solutions, we show that both the VP and the VV formulations are comparable in accuracy but their best performance corresponds to different NN architectures. The initial convergence rate is fast but the error eventually saturates to a plateau due to the dominance of the optimization error. For the turbulent channel flow, we show that NSFnets can sustain turbulence at , but due to expensive training we only consider part of the channel domain and enforce velocity boundary conditions on the subdomain boundaries provided by the DNS data base. We also perform a systematic study on the weights used in the loss function for balancing the data and physics components, and investigate a new way of computing the weights dynamically to accelerate training and enhance accuracy. In the last part, we demonstrate how NSFnets should be used in practice, namely for ill-posed problems with incomplete or noisy boundary conditions as well as for inverse problems. We obtain reasonably accurate solutions for such cases as well without the need to change the NSFnets and at the same computational cost as in the forward well-posed problems. As a result, we also present a simple example of transfer learning that will aid in accelerating the training of NSFnets for different parameter settings.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1973

Some effects of experimental error in fracture testing

Effects of experimental imprecision on the stress intensity factors calculated for various practical specimen types are shown. A general form equation for the stress intensity factor is presented and a general error equation is derived. The expected error in the stress intensity factor is given in terms of the precision levels of the basic experimental measurements and derivatives of the stress intensity calibration factor. Nine common fracture specimen types are considered, and the sensitivity of the various types to experimental error is illustrated. Some implications for fracture toughness testing and crack growth rate testing are discussed and methods of analysis are proposed to compensate for the effects of experimental error.

Orange, T. W.↗
NASA NTRS1985

A comparative study of spherical and flat-Earth geopotential modeling at satellite elevations

Flat-Earth and spherical-Earth geopotential modeling of crustal anomaly sources at satellite elevations are compared by computing gravity and scalar magnetic anomalies perpendicular to the strike of variably dimensioned rectangular prisms at altitudes of 150, 300, and 450 km. Results indicate that the error caused by the flat-Earth approximation is less than 10% in most geometric conditions. Generally, error increase with larger and wider anomaly sources at higher altitudes. For most crustal source modeling applications at conventional satellite altitudes, flat-Earth modeling can be justified and is numerically efficient.

Parrott, M. H.↗
NASA NTRSDate not supplied

Implementation of A New Microwave Scattering Database and A Forward Model for Active Microwave Sensors in CRTM

Radiative transfer models are extensively used for the assimilation of satellite observations into NWP models as well as retrieving geophysical products from satellite measurements. CRTM is a community model developed by NOAA JCSDA and widely used for different purposes requiring RT calculations. CRTM requires bulk scattering lookup tables in order to perform all-sky RT calculations. However, the current CRTM lookup tables for microwave frequencies were generated based on the Mie theory by assuming spherical frozen particles. The scattering lookup tables generated using the DDA technique has shown to largely improve the RT scattering calculations in the MW region. This presentation targets (i) the implementation and validation of a DDA database that was originally developed for the ARTS RT model into CRTM, and (ii) developing the CRTM active sensor module that takes advantage of the backscattering coefficients computed using the DDA method. The DDA database only provides single scattering properties of different habits, while CRTM requires bulk scattering properties. The CRTM cloud coefficients were previously generated based on the effective radius for representing the size of the particles. However, effective radius is neither measurable nor provided by the NWP models, thus need to be estimated from other geophysical variables such as water content. Therefore, in addition to calculating the CRTM bulk scattering properties from the DDA single scattering database, the CRTM was also largely modified to use cloud water content (kg.m-3), instead of effective radius, for performing the interpolation over size/mass of the particles. CRTM already requires water content as input, thus no extra variables are required for performing scattering calculations using the new ARTS DDA database. The CRTM scattering modules search for effective radius in cloud coefficient files and will use the cloud water content if the effective radius dimension is not found in the cloud coefficient files. Figure 1 shows the CRTM simulated brightness temperatures computed using different cloud coefficients versus ATMS observed values over Hurricane Irma on September 7, 2017 at 18:00 UTC. We used all the cloud water content values included in ERA5 with default CRTM/DDA habits for water, rain, snow, ice, hail, and graupel. ERA5 does not provide separate water content values for ice, hail, and graupel, thus the ice water content values were divided between ice, hail, and graupel clouds similar to what was explained in the previous section. In channels with a frequency lower than 90 GHz, emission from water and rain clouds can compensate for cloud scattering so that cloud contaminated Tbs are larger than corresponding clear sky Tbs. The DDA simulations for channels 1-7 largely perform better than the Mie simulations. The DDA simulations show a mix of small negative and positive simulated minus observed values, while the Mie results show large negative biases. The weighting functions for some of the ATMS temperature sounding channels (channels 9-15) peak mostly above the clouds, therefore the measured Tbs become less sensitive to clouds so that the results of both Mie and DDA become very similar. The Mie lookup tables generate excessive scattering for channel 16, but not enough scattering for the water vapor channels. In the specific case of Hurricane Maria, the DDA lookup tables do not generate enough scattering for channel 16, but the DDA results are much more consistent with observations for water vapor channels than for channel 16. It should be noted that the results may vary if we use other habits to represent snow, hail, and graupel in the DDA simulations. Although these results clearly show the advantage of the DDA database over the Mie dataset, different error sources such as error in the observations, displacement of clouds in the ERA5 reanalysis, and also lack of convective clouds or in general errors in the input atmospheric and cloud profiles contribute to the differences between the simulated and observed values. Aside from the improvements in the simulations, a major advantage of the new dataset is a large number of habits that can be used to tune the data assimilation systems to perform well in different weather conditions.

Isaac Moradi↗
NASA NTRS1973

Some effects of experimental error in fracture testing.

The purpose of this paper is to show the effects of experimental imprecision on the stress intensity factors calculated for various practical specimen types. A general form equation for the stress intensity factor is presented, and a general error equation is derived. The expected error in the stress intensity factor is given in terms of the precision levels of the basic experimental measurements and derivatives of the stress intensity calibration factor. Nine common fracture specimen types are considered, and the sensitivity of the various types to experimental error is illustrated. Some implications for fracture toughness testing and crack growth rate testing are discussed, and methods of analysis are proposed to compensate for the effects of experimental error.

Orange, T. W.↗
NASA NTRS1994

Comparison of Kalman filter and optimal smoother estimates of spacecraft attitude

Given a valid system model and adequate observability, a Kalman filter will converge toward the true system state with error statistics given by the estimated error covariance matrix. The errors generally do not continue to decrease. Rather, a balance is reached between the gain of information from new measurements and the loss of information during propagation. The errors can be further reduced, however, by a second pass through the data with an optimal smoother. This algorithm obtains the optimally weighted average of forward and backward propagating Kalman filters. It roughly halves the error covariance by including future as well as past measurements in each estimate. This paper investigates whether such benefits actually accrue in the application of an optimal smoother to spacecraft attitude determination. Tests are performed both with actual spacecraft data from the Extreme Ultraviolet Explorer (EUVE) and with simulated data for which the true state vector and noise statistics are exactly known.

Sedlak, J.↗
NASA NTRS1981

On the sensitivity of numerical weather prediction to remotely sensed marine surface wind data - A simulation study

The reported investigation has the objective to assess the potential impact on numerical weather prediction (NWP) of remotely sensed surface wind data. Other investigations conducted with similar objectives have not been satisfactory in connection with a use of procedures providing an unrealistic distribution of initial errors. In the current study, care has been taken to duplicate the actual distribution of information in the conventional observing system, thus shifting the emphasis from accuracy of the data to the data coverage. It is pointed out that this is an important consideration in assessing satellite observing systems since experience with sounder data has shown that improvements in forecasts due to satellite-derived information is due less to a general error reduction than to the ability to fill data-sparse regions. The reported study concentrates on the evaluation of the observing system simulation experimental design and on the assessment of the potential of remotely sensed marine surface wind data.

Cane, M. A.↗