Search NASA⌕ Search

SEARCH · Search NASA

Results for “algorithmic differentiation”

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 379 records · Page 21

A backward SDE method for uncertainty quantification in deep learning

Here, we develop a backward stochastic differential equation based probabilistic machine learning method, which formulates a class of stochastic neural networks as a stochastic optimal control problem. An efficient stochastic gradient descent algorithm is introduced with the gradient computed through a backward stochastic differential equation. Convergence analysis for stochastic gradient descent optimization and numerical experiments for applications of stochastic neural networks are carried out to validate our methodology in both theory and performance.

97 MATHEMATICS AND COMPUTING↗

Adapting Wave-front Algorithms to Efficiently Utilize Systems with Deep Communication Hierarchies

Large-scale systems increasingly exhibit a differential between intra-chip and inter-chip communication performance especially in hybrid systems using accelerators. Processorcores on the same socket are able to communicate at lower latencies, and with higher bandwidths, than cores on different sockets either within the same node or between nodes. A key challenge is to efficiently use this communication hierarchy and hence optimize performance. We consider here the class of applications that contains wavefront processing. In these applications data can only be processed after their upstream neighbors have been processed. Similar dependencies result between processors in which communication is required to pass boundary data downstream and whose cost is typically impacted by the slowest communication channel in use. In this work we develop a novel hierarchical wave-front approach that reduces the use of slower communications in the hierarchy but at the cost of additional steps in the parallel computation and higher use of on-chip communications. This tradeoff is explored using a performance model. An implementation using the Reverse-acceleration programming model on the petascale Roadrunner system demonstrates a 27% performance improvement at full system-scale on a kernel application. The approach is generally applicable to large-scale multi-core and accelerated systems where a differential in system communication performance exists.

43 PARTICLE ACCELERATORS↗

Parallel adaptive wavelet collocation method for PDEs

A parallel adaptive wavelet collocation method for solving a large class of Partial Differential Equations is presented. The parallelization is achieved by developing an asynchronous parallel wavelet transform, which allows one to perform parallel wavelet transform and derivative calculations with only one data synchronization at the highest level of resolution. The data are stored using tree-like structure with tree roots starting at a priori defined level of resolution. Both static and dynamic domain partitioning approaches are developed. For the dynamic domain partitioning, trees are considered to be the minimum quanta of data to be migrated between the processes. This allows fully automated and efficient handling of non-simply connected partitioning of a computational domain. Dynamic load balancing is achieved via domain repartitioning during the grid adaptation step and reassigning trees to the appropriate processes to ensure approximately the same number of grid points on each process. The parallel efficiency of the approach is discussed based on parallel adaptive wavelet-based Coherent Vortex Simulations of homogeneous turbulence with linear forcing at effective non-adaptive resolutions up to 2048{sup 3} using as many as 2048 CPU cores.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

A unified view of hierarchy approach and formula of differentiation

The stochastic differential equation is a powerful tool for describing the dynamics of a dissipative system in which noise characterizes the influence of the environment. For the Ornstein-Uhlenbeck noise, both the formula of differentiation and the hierarchy approach provide efficient numerical simulations, with the stochastic differential equation transformed into a set of coupled, linear ordinary differential equations. We show that while these two deterministic schemes result in different sets of equations, they can be regarded as two representations of an underlying linear-dynamics. Moreover, by manipulating the involved Ornstein-Uhlenbeck noise, we propose a unified algorithm that may reduce to the hierarchy approach or the formula of differentiation in different limits. We further analyze the numerical performance of this algorithm and find that the hierarchy approach appears to be more efficient for our numerical model studies.

Yan, Yun-An (ORCID:000000031943459X)↗

Assessing three closed-loop learning algorithms by searching for high-quality quantum control pulses

Designing a high-quality control is crucial for reliable quantum computation. Among the existing approaches, closed-loop leaning control is an effective choice. Its efficiency depends on the learning algorithm employed, thus deserving algorithmic comparisons for its practical applications. Here we assess three representative learning algorithms, including GRadient Ascent Pulse Engineering (GRAPE), improved Nelder-Mead (NMplus), and Differential Evolution (DE), by searching for high-quality control pulses to prepare the Bell state. We first implement each algorithm experimentally in a nuclear magnetic resonance system and then conduct a numerical study considering the impact of some possible significant experimental uncertainties. The experiments report the successful preparation of the high-fidelity target state by the three algorithms, while NMplus converges fastest, and these results coincide with the numerical simulations when potential uncertainties are negligible. However, under certain significant uncertainties, these algorithms possess distinct performance with respect to their resulting precision and efficiency, and DE shows the best robustness. Finally, this study provides insight to aid in the practical application of different closed-loop learning algorithms in realistic physical scenarios.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Numerical simulation of photoexcited polaron states in water

We consider the dynamic polaron model of the hydrated electron state on the basis of a system of three nonlinear partial differential equations with appropriate initial and boundary conditions. A parallel numerical algorithm for the numerical solution of this system has been developed. Its effectiveness has been tested on a few multi-processor systems. A numerical simulation of the polaron states formation in water under the action of the ultraviolet range laser irradiation has been performed. The numerical results are shown to be in a reasonable agreement with experimental data and theoretical predictions.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Differentiable Predictions for Large Scale Structure with SHAMNet

In simulation-based models of the galaxy-halo connection, theoretical predictions for galaxy clustering and lensing are typically made based on Monte Carlo realizations of a mock universe. In this paper, we use Subhalo Abundance Matching (SHAM) as a toy model to introduce an alternative to stochastic predictions based on mock population, demonstrating how to make simulation-based predictions for clustering and lensing that are both exact and differentiable with respect to the parameters of the model. Conventional implementations of SHAM are based on iterative algorithms such as Richardson-Lucy deconvolution; here we use the JAX library for automatic differentiation to train SHAMNet, a neural network that accurately approximates the stellar-to-halo mass relation (SMHM) defined by abundance matching. In our approach to making differentiable predictions of large-scale structure, we map parameterized PDFs onto each simulated halo, and calculate gradients of summary statistics of the galaxy distribution by using autodiff to propagate the gradients of the SMHM through the statistical estimators used to measure one- and two-point functions. Our techniques are quite general, and we conclude with an overview of how they can be applied in tandem with more complex, higher-dimensional models, creating the capability to make differentiable predictions for the multi-wavelength universe of galaxies.

79 ASTRONOMY AND ASTROPHYSICS↗

An ensemble Kalman filter for statistical estimation of physics constrained nonlinear regression models

A central issue in contemporary science is the development of nonlinear data driven statistical–dynamical models for time series of noisy partial observations from nature or a complex model. It has been established recently that ad-hoc quadratic multi-level regression models can have finite-time blow-up of statistical solutions and/or pathological behavior of their invariant measure. Recently, a new class of physics constrained nonlinear regression models were developed to ameliorate this pathological behavior. Here a new finite ensemble Kalman filtering algorithm is developed for estimating the state, the linear and nonlinear model coefficients, the model and the observation noise covariances from available partial noisy observations of the state. Several stringent tests and applications of the method are developed here. In the most complex application, the perfect model has 57 degrees of freedom involving a zonal (east–west) jet, two topographic Rossby waves, and 54 nonlinearly interacting Rossby waves; the perfect model has significant non-Gaussian statistics in the zonal jet with blocked and unblocked regimes and a non-Gaussian skewed distribution due to interaction with the other 56 modes. We only observe the zonal jet contaminated by noise and apply the ensemble filter algorithm for estimation. Numerically, we find that a three dimensional nonlinear stochastic model with one level of memory mimics the statistical effect of the other 56 modes on the zonal jet in an accurate fashion, including the skew non-Gaussian distribution and autocorrelation decay. On the other hand, a similar stochastic model with zero memory levels fails to capture the crucial non-Gaussian behavior of the zonal jet from the perfect 57-mode model.

97 MATHEMATICS AND COMPUTING↗

THE MAGNETOHYDRODYNAMIC RESPONSE OF LIQUID OXYGEN: EXPERIMENTATION AND SIMULATION

Experimental and theoretical studies have been conducted to establish the basic understanding and predictive capability for the dynamics of a liquid oxygen (LOX) slug subjected to magnetic fields within a solenoid. The electrically-pulsed solenoids around a 1.9 mm ID quartz tube were capable of producing up to 1.1 T when immersed in liquid nitrogen. The slug dynamics were measured by pressure changes in a closed volume on both sides of the slug. A theoretical model was developed which balances the magnetic, viscous, and pressure forces into a single equation of motion. The model was applied to a one-dimensional discretized algorithm that solved the coupled multiphysics problem of the Navier-Stokes and Maxwell's equations. The simulation and experimental results established LOX as a good candidate in a magnetic fluid system without moving parts for cryogenic applications.

36 MATERIALS SCIENCE↗

Measurement of photonuclear jet production in ultraperipheral Pb + Pb collisions at $\sqrt{s_{NN}}$ = 5.02 TeV with the ATLAS detector

In ultrarelativistic heavy ion collisions at the LHC, each nucleus acts a sources of high-energy real photons that can scatter off the opposing nucleus in ultraperipheral photonuclear (𝛾 + 𝐴) collisions. Hard scattering processes initiated by the photons in such collisions provide a novel method for probing nuclear parton distributions in a kinematic region not easily accessible to other measurements. ATLAS has measured production of dijet and multijet final states in ultraperipheral Pb + Pb collisions at $\sqrt{s_{NN}}$ = 5.02 TeV using a dataset recorded in 2018 with an integrated luminosity of 1.72 nb −1 . Photonuclear final states are selected by requiring a rapidity gap in the photon direction; this selects events where one of the outgoing nuclei remains intact. Jets are reconstructed using the anti-𝑘 t algorithm with radius parameter, 𝑅 = 0.4. Triple-differential cross sections, unfolded for detector response, are measured and presented using two sets of kinematic variables. The first set consists of the total transverse momentum (𝐻 T ), rapidity, and mass of the jet system. The second set uses 𝐻 T and particle-level nuclear and photon parton momentum fractions, 𝑥 A and 𝑧 𝛾 , respectively. The results are compared with leading-order perturbative QCD calculations of photonuclear jet production cross sections, where all leading order predictions using existing fits fall below the data in the shadowing region. More detailed theoretical comparisons will allow these results to strongly constrain nuclear parton distributions, and these data provide results from the LHC directly comparable to early physics results at the planned Electron-Ion Collider.

Parton distribution functions↗

GPU-based fast cone beam CT reconstruction from undersampled and noisy projection data via total variation

Purpose: Cone-beam CT (CBCT) plays an important role in image guided radiation therapy (IGRT). However, the large radiation dose from serial CBCT scans in most IGRT procedures raises a clinical concern, especially for pediatric patients who are essentially excluded from receiving IGRT for this reason. The goal of this work is to develop a fast GPU-based algorithm to reconstruct CBCT from undersampled and noisy projection data so as to lower the imaging dose. Methods: The CBCT is reconstructed by minimizing an energy functional consisting of a data fidelity term and a total variation regularization term. The authors developed a GPU-friendly version of the forward-backward splitting algorithm to solve this model. A multigrid technique is also employed. Results: It is found that 20-40 x-ray projections are sufficient to reconstruct images with satisfactory quality for IGRT. The reconstruction time ranges from 77 to 130 s on an NVIDIA Tesla C1060 (NVIDIA, Santa Clara, CA) GPU card, depending on the number of projections used, which is estimated about 100 times faster than similar iterative reconstruction approaches. Moreover, phantom studies indicate that the algorithm enables the CBCT to be reconstructed under a scanning protocol with as low as 0.1 mA s/projection. Comparing with currently widely used full-fan head and neck scanning protocol of {approx}360 projections with 0.4 mA s/projection, it is estimated that an overall 36-72 times dose reduction has been achieved in our fast CBCT reconstruction algorithm. Conclusions: This work indicates that the developed GPU-based CBCT reconstruction algorithm is capable of lowering imaging dose considerably. The high computation efficiency in this algorithm makes the iterative CBCT reconstruction approach applicable in real clinical environments.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

SU-E-T-183: Clinical Quality Assurance Workflow for Dynamic Tumor Tracking Radiation Dose Delivery

Purpose: One of the most important aspects of implementation of new treatment modalities is an ‘end-to-end’ verification of the treatment process. Radiation treatment based on dynamic tracking of a tumor is highly patient-specific, therefore, special attention should be paid to quality assurance of the treatment delivery. Our goal was to design the clinical workflow that ensures accurate delivery of the planned dose using the Dynamic Target Tracking option of VeroTM (BrainLab,MHI) linac. Methods: A patient simulation is designed to include a pre-treatment session to verify whether the system can reliably track the motion of the implanted marker and build the 4D model of the target motion. The external surrogate and target motion patterns are recorded in the ExactracTM log files. In this work, a spectrum of custom marker and external surrogate motion trajectories closely resembling the patient specific motion patterns was used. 1mm thick/11mm long VisicoilTM marker was placed 15 and 20mm from the center of the spherical tissue equivalent target (centroid to centroid distance) in the 4D motion phantom (CIRSTM). 3D conformal (3 mm block margin) SBRT plans were delivered to 2 moving targets in the phantom: 1) 20mm diameter target that allows ion chamber dose measurement and 2) 25mm target that allows using film to measure CAX dose (GafchromicTM EBT3 used). The measured dose was compared to the iPlanTM TPS results using MonteCarlo algorithm (1% variance, Dose-to-water). Results: On average, film shows 98.9% pass using gamma criterion for 2% and 2mm DTA, 94.3% match for 2% and 1 mm DTA, 98% pass for 1% and 2 mm DTA however only 88% points passing for 1% and 1 mm DTA. Ion chamber measurements agreed with the calculation within 1.5%. Conclusion: The clinical QA workflow was designed for SBRT delivery using real-time tumor tracking on VeroTM linac.

60 APPLIED LIFE SCIENCES↗

SU-F-P-39: End-To-End Validation of a 6 MV High Dose Rate Photon Beam, Configured for Eclipse AAA Algorithm Using Golden Beam Data, for SBRT Treatments Using RapidArc

Purpose: To use end-to-end testing to validate a 6 MV high dose rate photon beam, configured for Eclipse AAA algorithm using Golden Beam Data (GBD), for SBRT treatments using RapidArc. Methods: Beam data was configured for Varian Eclipse AAA algorithm using the GBD provided by the vendor. Transverse and diagonals dose profiles, PDDs and output factors down to a field size of 2×2 cm2 were measured on a Varian Trilogy Linac and compared with GBD library using 2% 2mm 1D gamma analysis. The MLC transmission factor and dosimetric leaf gap were determined to characterize the MLC in Eclipse. Mechanical and dosimetric tests were performed combining different gantry rotation speeds, dose rates and leaf speeds to evaluate the delivery system performance according to VMAT accuracy requirements. An end-to-end test was implemented planning several SBRT RapidArc treatments on a CIRS 002LFC IMRT Thorax Phantom. The CT scanner calibration curve was acquired and loaded in Eclipse. PTW 31013 ionization chamber was used with Keithley 35617EBS electrometer for absolute point dose measurements in water and lung equivalent inserts. TPS calculated planar dose distributions were compared to those measured using EPID and MapCheck, as an independent verification method. Results were evaluated with gamma criteria of 2% dose difference and 2mm DTA for 95% of points. Results: GBD set vs. measured data passed 2% 2mm 1D gamma analysis even for small fields. Machine performance tests show results are independent of machine delivery configuration, as expected. Absolute point dosimetry comparison resulted within 4% for the worst case scenario in lung. Over 97% of the points evaluated in dose distributions passed gamma index analysis. Conclusion: Eclipse AAA algorithm configuration of the 6 MV high dose rate photon beam using GBD proved efficient. End-to-end test dose calculation results indicate it can be used clinically for SBRT using RapidArc.

60 APPLIED LIFE SCIENCES↗

A pencil beam dose calculation model for CyberKnife system

Purpose: CyberKnife system is initially equipped with fixed circular cones for stereotactic radiosurgery. Two dose calculation algorithms, Ray-Tracing and Monte Carlo, are available in the supplied treatment planning system. A multileaf collimator system was recently introduced in the latest generation of system, capable of arbitrarily shaped treatment field. The purpose of this study is to develop a model based dose calculation algorithm to better handle the lateral scatter in an irregularly shaped small field for the CyberKnife system. Methods: A pencil beam dose calculation algorithm widely used in linac based treatment planning system was modified. The kernel parameters and intensity profile were systematically determined by fitting to the commissioning data. The model was tuned using only a subset of measured data (4 out of 12 cones) and applied to all fixed circular cones for evaluation. The root mean square (RMS) of the difference between the measured and calculated tissue-phantom-ratios (TPRs) and off-center-ratio (OCR) was compared. Three cone size correction techniques were developed to better fit the OCRs at the penumbra region, which are further evaluated by the output factors (OFs). The pencil beam model was further validated against measurement data on the variable dodecagon-shaped Iris collimators and a half-beam blocked field. Comparison with Ray-Tracing and Monte Carlo methods was also performed on a lung SBRT case. Results: The RMS between the measured and calculated TPRs is 0.7% averaged for all cones, with the descending region at 0.5%. The RMSs of OCR at infield and outfield regions are both at 0.5%. The distance to agreement (DTA) at the OCR penumbra region is 0.2 mm. All three cone size correction models achieve the same improvement in OCR agreement, with the effective source shift model (SSM) preferred, due to their ability to predict more accurately the OF variations with the source to axis distance (SAD). In noncircular field validation, the pencil beam calculated results agreed well with the film measurement of both Iris collimators and the half-beam blocked field, fared much better than the Ray-Tracing calculation. Conclusions: The authors have developed a pencil beam dose calculation model for the CyberKnife system. The dose calculation accuracy is better than the standard linac based system because the model parameters were specifically tuned to the CyberKnife system and geometry correction factors. The model handles better the lateral scatter and has the potential to be used for the irregularly shaped fields. Comprehensive validations on MLC equipped system are necessary for its clinical implementation. It is reasonably fast enough to be used during plan optimization.

60 APPLIED LIFE SCIENCES↗

Machine-Learning-Based Multiscale Methods for 3D Modelling of Granular Materials by Incorporating History-Dependent State Variables

Over the past decades, the prevalence of machine learning (ML) methods has made the development of ML-based constitutive models for granular materials undoubtedly a popular subject. Numerous studies have been made to feature the loading path or history-dependent stress-strain response of granular media using neural networks. In this work, a novel finite element method (FEM)–ML multiscale approach was developed by incorporating internal variables to improve the simulation accuracy of 3D history-dependent granular materials for the first time. To this end, a surrogate constitutive model based on the single-step-based multi-layer perceptron (MLP) neural network was used to replace representative volume element (RVE) simulations conducted by the discrete element method (DEM) in the multiscale FEM–DEM approach. Although the prediction principle of the MLP aligns with the FEM algorithm, artificially added internal variables are required to differentiate the loading history. To address this issue, history variables associated with the Frobenius norm are proposed to be fed into the MLP coupled with the strain tensor to extract the history-dependent behaviour of granular assemblies. The developed FEM–ML approach was demonstrated in 3D conventional triaxial compression (CTC) simulations. Compared to the multiscale FEM–DEM approach, the proposed FEM–ML method exhibits a significantly improved computational efficiency.

granular materials↗

Automated Thermal Image Processing for Detection and Classification of Birds and Bats (FY2012 Annual Report)

Surveying wildlife at risk from offshore wind energy development is difficult and expensive. Infrared video can be used to record birds and bats that pass through the camera view, but it is also time consuming and expensive to review video and determine what was recorded. We proposed to conduct algorithm and software development to identify and to differentiate thermally detected targets of interest that would allow automated processing of thermal image data to enumerate birds, bats, and insects. During FY2012 we developed computer code within MATLAB to identify objects recorded in video and extract attribute information that describes the objects recorded. We tested the efficiency of track identification using observer-based counts of tracks within segments of sample video. We examined object attributes, modeled the effects of random variability on attributes, and produced data smoothing techniques to limit random variation within attribute data. We also began drafting and testing methodology to identify objects recorded on video. We also recorded approximately 10 hours of infrared video of various marine birds, passerine birds, and bats near the Pacific Northwest National Laboratory (PNNL) Marine Sciences Laboratory (MSL) at Sequim, Washington. A total of 6 hours of bird video was captured overlooking Sequim Bay over a series of weeks. An additional 2 hours of video of birds was also captured during two weeks overlooking Dungeness Bay within the Strait of Juan de Fuca. Bats and passerine birds (swallows) were also recorded at dusk on the MSL campus during nine evenings. An observer noted the identity of objects viewed through the camera concurrently with recording. These video files will provide the information necessary to produce and test software developed during FY2013. The annotation will also form the basis for creation of a method to reliably identify recorded objects.

Bats↗