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 1,405 records · Page 78

Structures via Reasoning - Applying AI to Cryo Electron Microscopy to Reveal Structural Variability

There have been breakthroughs of latest cryo electron microscopy (cryo-EM) data analysis algorithms to classify cryo-EM image data. However, most of these cryo-EM reconstruction methods have focused on classifying distinctly different biomolecule structures. Here, we present our approaches of deep learning to differentiate homologous structures that are distinguishable only with inner morphological differences. We succeeded supervised classification of these subtly different homologues. However, we could not differentiate them with unsupervised methods. Here we discuss what further approaches are likely needed for successful unsupervised classification.

47 OTHER INSTRUMENTATION↗

Labeled trees and the efficient computation of derivations

The effective parallel symbolic computation of operators under composition is discussed. Examples include differential operators under composition and vector fields under the Lie bracket. Data structures consisting of formal linear combinations of rooted labeled trees are discussed. A multiplication on rooted labeled trees is defined, thereby making the set of these data structures into an associative algebra. An algebra homomorphism is defined from the original algebra of operators into this algebra of trees. An algebra homomorphism from the algebra of trees into the algebra of differential operators is then described. The cancellation which occurs when noncommuting operators are expressed in terms of commuting ones occurs naturally when the operators are represented using this data structure. This leads to an algorithm which, for operators which are derivations, speeds up the computation exponentially in the degree of the operator. It is shown that the algebra of trees leads naturally to a parallel version of the algorithm.

Grossman, Robert↗

TPSAS-NF1676L-23557-DND

The Cloud-Aerosol LIdar with Orthogonal Polarization (CALIOP) measures both parallel and perpendicular backscatter of linearly polarized 532 nm light from cloud particles in the atmosphere. The perpendicular component of the return signal allows differentiation between cloud layers consisting of spherical water droplets from those with non-spherical ice particles. A two-way distribution of layer-integrated 532 nm total backscatter and the layer-integrated 532 nm depolarization ratio further differentiates between water cloud layers with significant multiple scattering, horizontally-oriented ice particles that cause specular reflections, and randomly-oriented ice particles, as described in Hu et al., 2009. The cloud phase algorithm functions well for the majority of clouds, but there have been some recent changes and improvements that are described in this poster. The impact of the CALIOP viewing angle change from near-nadir (0.3 degrees) to slightly tilted (3.0 degrees) on the optical properties of the various CALIOP cloud particle phases is also shown. A major improvement to the phase algorithm is the limitation of a spatial coherence test for horizontally-oriented ice to cloud layers that have sufficient integrated attenuated backscatter to likely contain HOI. This change reduces a "striping" effect caused by noisiness in the spatial coherence test for thin cloud layers. Additional temperature tests are considered, and the regional and vertical distribution of cloud layers with the various CALIOP cloud phases is shown. Only one cloud particle phase is allowed for each 5 km portion of a cloud layer that is considered by the cloud phase algorithm, so we use in situ aircraft and CloudSat observations to determine how the CALIOP cloud phase algorithm is treating optically and thermally thick clouds, which may be mixed-phase.

Melody Avery↗

Differentiable Preisach Modeling for Characterization and Optimization of Particle Accelerator Systems with Hysteresis

Future improvements in particle accelerator performance are predicated on increasingly accurate online modeling of accelerators. Hysteresis effects in magnetic, mechanical, and material components of accelerators are often neglected in online accelerator models used to inform control algorithms, even though reproducibility errors from systems exhibiting hysteresis are not negligible in high precision accelerators. Here, we combine the classical Preisach model of hysteresis with machine learning techniques to efficiently create nonparametric, high-fidelity models of arbitrary systems exhibiting hysteresis. We experimentally demonstrate how these methods can be used in situ, where a hysteresis model of an accelerator magnet is combined with a Bayesian statistical model of the beam response, allowing characterization of magnetic hysteresis solely from beam-based measurements. Finally, we explore how using these joint hysteresis-Bayesian statistical models allows us to overcome optimization performance limitations that arise when hysteresis effects are ignored.

43 PARTICLE ACCELERATORS↗

TU-C-12A-12: Differentiating Bone Lesions and Degenerative Joint Disease in NaF PET/CT Scans Using Machine Learning

Purpose: [F-18]NaF PET can be used to image bone metastases; however, tracer uptake in degenerative joint disease (DJD) often appears similar to metastases. This study aims to develop and compare different machine learning algorithms to automatically identify regions of [F-18]NaF scans that correspond to DJD. Methods: 10 metastatic prostate cancer patients received whole body [F-18]NaF PET/CT scans prior to treatment. Image segmentation resulted in 852 ROIs, 69 of which were identified by a nuclear medicine physician as DJD. For all ROIs, various PET and CT textural features were computed. ROIs were divided into training and testing sets used to train eight different machine learning classifiers. Classifiers were evaluated based on receiver operating characteristics area under the curve (AUC), sensitivity, specificity, and positive predictive value (PPV). We also assessed the added value of including CT features in addition to PET features for training classifiers. Results: The training set consisted of 37 DJD ROIs with 475 non-DJD ROIs, and the testing set consisted of 32 DJD ROIs with 308 non-DJD ROIs. Of all classifiers, generalized linear models (GLM), decision forests (DF), and support vector machines (SVM) had the best performance. AUCs of GLM (0.929), DF (0.921), and SVM (0.889) were significantly higher than the other models (p<0.001). GLM and DF, overall, had the best sensitivity, specificity, and PPV, and gave a significantly better performance (p<0.01) than all other models. PET/CT GLM classifiers had higher AUC than just PET or just CT. GLMs built using PET/CT information had superior or comparable sensitivities, specificities and PPVs to just PET or just CT. Conclusion: Machine learning algorithms trained with PET/CT features were able to identify some cases of DJD. GLM outperformed the other classification algorithms. Using PET and CT information together was shown to be superior to using PET or CT features alone. Research supported by the Prostate Cancer Foundation.

60 APPLIED LIFE SCIENCES↗

Implicit Formulation of Muscle Dynamics in OpenSim

Astronauts lose bone and muscle mass during spaceflight. Exercise countermeasure is the primary method for counteracting bone and muscle mass loss in space. New spacecraft exercise device concepts are currently being developed for the NASAs new crew exploration vehicle. The NASA Digital Astronaut Project (DAP) uses computational modeling to help determine if the new exercise devices will be effective as countermeasures. The NASA Digital Astronaut Project is developing the ability to utilize predictive simulation to provide insight into the change in kinematics and kinetics with a change in device and gravitational environment (1-g versus 0-g). For example, in space exercise the subject's body weight is applied in addition to the loads prescribed for musculoskeletal maintenance. How and where these loads are applied obviously directly impacts bone and tissue loads. Additionally, due to space vehicle structural requirements, exercise devices are often placed on vibration isolation systems. This changes the apparent impedance or stiffness of the device as seen by the user. Data collection under these conditions is often impractical and limited. Predictive modeling provides a means to have a virtual subject to test hypotheses. Predictive simulation provides a virtual subject for which we are able to perform studies such as sensitivity to device loading and vibration isolation without the need for laboratory kinematic or kinetic test data.Direct Collocation optimization provides an efficient means to perform task based optimization and predictive modeling. It is relatively straight forward to structure a physical exercise task in a Direct Collocation mathematical formulation: perform a motion such that you start at an initial pose, achieve a given amount of deflection i.e a squat, return to the initial pose, and minimize muscle activation cost. Direct Collocation is advantageous in that it does not require numerical integration to evaluate the objective function. Instead, the system dynamics are transformed to discrete time and the optimizer is constrained such that the solution is not considered to be a valid unless the dynamic equations are satisfied at all time points. The simulation and optimization are effectively done simultaneously. Due to the implicit integration, time steps can be more coarse than in a differential equation solver. In a gait scenario this means that that the model constraints and cost function are evaluated at 100 nodes in the gait cycle versus 10,000 integration steps in a variable-step forward dynamic simulation. Furthermore, no time is wasted on accurate simulations of movements that are far from the optimum. Constrained optimization algorithms require a Jacobian matrix that contains the partial derivatives of each of the dynamic constraints with respect to of each of the state and control variables at all time points. This is a large but sparse matrix. An implicit dynamics formulation requires computation of the dynamic residuals f as a function of the states x and their derivatives, and controls u:f(x, dxdt, u) 0If the dynamics of musculoskeletal system are formulated implicitly, the Jacobian elements are often available analytically, eliminating the need for numerical differentiation; this is obviously computationally advantageous. Additionally, implicit formulation of musculoskeletal dynamics do not suffer from singularities from low mass bodies, zero muscle activation, or other stiff system or

physical exercise↗

Direct simulation of compressible turbulence

Several direct simulations of 3-D homogeneous, compressible turbulence are presented with emphasis on the differences with incompressible turbulent simulations. A fully spectral collocation algorithm, periodic in all directions coupled with a 3rd order Runge-Kutta time discretization scheme is sufficient to produce well-resolved flows at Taylor Reynolds numbers below 40 on grids of 128x128x128. A Helmholtz decomposition of velocity is useful to differentiate between the purely compressible effects and those effects solely due to vorticity production. In the context of homogeneous flows, this decomposition in unique. Time-dependent energy and dissipation spectra of the compressible and solenoidal velocity components indicate the presence of localized small scale structures. These structures are strongly a function of the initial conditions. Researchers concentrate on a regime characterized by very small fluctuating Mach numbers Ma (on the order of 0.03) and density and temperature fluctuations much greater than sq Ma. This leads to a state in which more than 70 percent of the kinetic energy is contained in the so-called compressible component of the velocity. Furthermore, these conditions lead to the formation of curved weak shocks (or shocklets) which travel at approximately the sound speed across the physical domain. Various terms in the vorticity and divergence of velocity production equations are plotted versus time to gain some understanding of how small scales are actually formed. Possible links with Burger turbulence are examined. To visualize better the dynamics of the flow, new graphic visualization techniques have been developed. The 3-D structure of the shocks are visualized with the help of volume rendering algorithms developed in-house. A combination of stereographic projection and animation greatly increase the number of visual cues necessary to properly interpret the complex flow.

Zang, T. A.↗

TH-EF-207A-07: An Integrated X-Ray/bioluminescence Tomography System for Radiation Guidance and Tumor Evaluation

Purpose: CT is not able to differentiate tumors from surrounding soft tissue. This study is to develop a bioluminescence tomography (BLT) system that is integrated onto our previously developed CT guided small animal arc radiation treatment system (iSMAART) to guide radiation, monitor tumor growth and evaluate therapeutic response. Methods: The BLT system employs a CCD camera coupled with a high speed lens, and is aligned orthogonally to the x-ray beam central axis. The two imaging modalities, CT and BLT, are physically registered through geometrical calibration. The CT anatomy provides an accurate contour of animal surface which is used to construct 3D mesh for BLT reconstruction. Bioluminescence projections are captured from multiple angles, once every 45 degree rotation. The diffusion equation based on analytical Kirchhoff approximation is adopted to model the photon propagation in tissues. A discrete cosine transform based reweighted L1-norm regularization (DCT-re-L1) algorithm is used for BLT reconstruction. Experiments are conducted on a mouse orthotopic prostate tumor model (n=12) to evaluate the BLT performance, in terms of its robustness and accuracy in locating and quantifying the bioluminescent tumor cells. Iodinated contrast agent was injected intravenously to delineate the tumor in CT. The tumor location and volume obtained from CT also serve as a benchmark against BLT. Results: With our cutting edge reconstruction algorithm, BLT is able to accurately reconstruct the orthotopic prostate tumors. The tumor center of mass in BLT is within 0.5 mm radial distance of that in CT. The tumor volume in BLT is significantly correlated with that in CT (R2 = 0.81). Conclusion: The BLT can differentiate, localize and quantify tumors. Together with CT, BLT will provide precision radiation guidance and reliable treatment assessment in preclinical cancer research.

60 APPLIED LIFE SCIENCES↗

Numerical simulation of a thermodynamically consistent four‐species tumor growth model

We develop a thermodynamically consistent four‐species model of tumor growth on the basis of the continuum theory of mixtures. Unique to this model is the incorporation of nutrient within the mixture as opposed to being modeled with an auxiliary reaction‐diffusion equation. The formulation involves systems of highly nonlinear partial differential equations of surface effects through diffuse‐interface models. A mixed finite element spatial discretization is developed and implemented to provide numerical results demonstrating the range of solutions this model can produce. A time‐stepping algorithm is then presented for this system, which is shown to be first order accurate and energy gradient stable. The results of an array of numerical experiments are presented, which demonstrate a wide range of solutions produced by various choices of model parameters.

97 MATHEMATICS AND COMPUTING↗

Learning Optimal Multigrid Smoothers via Neural Networks

Multigrid methods are one of the most efficient techniques for solving large sparse linear systems arising from partial differential equations (PDEs) and graph Laplacians from machine learning applications. One of the key components of multigrid is smoothing, which aims at reducing high-frequency errors on each grid level. However, finding optimal smoothing algorithms is problem-dependent and can impose challenges for many problems. In this paper, we propose an efficient adaptive framework for learning optimized smoothers from operator stencils in the form of convolutional neural networks (CNNs). Here, the CNNs are trained on small-scale problems from a given type of PDEs based on a supervised loss function derived from multigrid convergence theories and can be applied to large-scale problems of the same class of PDEs. Numerical results on anisotropic rotated Laplacian problems and variable coefficient diffusion problems demonstrate improved convergence rates and solution time compared with classical hand-crafted relaxation methods.

97 MATHEMATICS AND COMPUTING↗

hIPPYlib-MUQ: A Bayesian Inference Software Framework for Integration of Data with Complex Predictive Models under Uncertainty

Bayesian inference provides a systematic framework for integration of data with mathematical models to quantify the uncertainty in the solution of the inverse problem. However, the solution of Bayesian inverse problems governed by complex forward models described by partial differential equations (PDEs) remains prohibitive with black-box Markov chain Monte Carlo (MCMC) methods. We present hIPPYlib-MUQ, an extensible and scalable software framework that contains implementations of state-of-the art algorithms aimed to overcome the challenges of high-dimensional, PDE-constrained Bayesian inverse problems. These algorithms accelerate MCMC sampling by exploiting the geometry and intrinsic low-dimensionality of parameter space via derivative information and low rank approximation. The software integrates two complementary open-source software packages, hIPPYlib and MUQ. hIPPYlib solves PDE-constrained inverse problems using automatically-generated adjoint-based derivatives, but it lacks full Bayesian capabilities. MUQ provides a spectrum of powerful Bayesian inversion models and algorithms, but expects forward models to come equipped with gradients and Hessians to permit large-scale solution. By combining these two complementary libraries, we created a robust, scalable, and efficient software framework that realizes the benefits of each and allows us to tackle complex large-scale Bayesian inverse problems across a broad spectrum of scientific and engineering disciplines. To illustrate the capabilities of hIPPYlib-MUQ, we present a comparison of a number of MCMC methods available in the integrated software on several high-dimensional Bayesian inverse problems. These include problems characterized by both linear and nonlinear PDEs, various noise models, and different parameter dimensions. The results demonstrate that large (~ 50×) speedups over conventional black box and gradient-based MCMC algorithms can be obtained by exploiting Hessian information (from the log-posterior), underscoring the power of the integrated hIPPYlib-MUQ framework.

97 MATHEMATICS AND COMPUTING↗

TAO 3.10 Users Manual

The Toolkit for Advanced Optimization (TAO) focuses on the development of algorithms and software for the solution of large-scale optimization problems on high-performance architectures. Areas of interest include unconstrained and bound-constrained optimization, nonlinear least squares problems, optimization problems with partial differential equation constraints, and variational inequalities and complementarity constraints.

97 MATHEMATICS AND COMPUTING↗

Early Career Award: An Enabling Computational Framework for Uncertainty Assimilation and Propagation in Complex PDE Systems: Sparse and Low-rank Techniques (Final Report)

This project aims to develop a set of new theories, algorithms, and computational tools that enable ultra-scalable representation and propagation of high-dimensional uncertainties to circumvent the issue of curse-of-dimensionality. The research components are motivated by the fact that highly scalable algorithms are needed to achieve exascale performance for uncertainty quantification and predictive simulation at extreme scales. Although these developments are generic and applicable to a wide range of high-dimensional, complex, stochastic, partial-differential-equation (PDE) systems, this research effort will focus on predictive simulation of high-energy-density lithium batteries. To create knowledge about the precise failure mechanisms of these batteries and ultimately predict their performance and reliability, a truly data-driven, multi-scale, multi-physics framework is needed that allows studying (i) dominant sources of uncertainty in the governing physical models at all scales of relevance and (ii) the interface between the interchanging physics and scales. The impact of the planned mathematical and numerical advancements in prediction of high-dimensional uncertain systems will be broad across areas of engineering and science such as climate modeling, fusion energy, combustion, and nanoscience, among others.

97 MATHEMATICS AND COMPUTING↗

Bilevel parameter optimization for learning nonlocal image denoising models

We propose a bilevel optimization approach for the estimation of parameters in nonlocal image denoising models. The parameters we consider are both the space-dependent fidelity weight and weights within the kernel of the nonlocal operator. In both cases we investigate the differentiability of the solution operator in function spaces and derive a first order optimality system that characterizes local minima. For the numerical solution of the problems, we propose a second-order trust-region algorithm in combination with a finite element discretization of the nonlocal denoising models and we introduce a computational strategy for the solution of the resulting dense linear systems. Several experiments illustrate the applicability and effectiveness of our approach.

97 MATHEMATICS AND COMPUTING↗

TAO Users Manual (Rev. 3.15)

The Toolkit for Advanced Optimization (TAO) focuses on the development of algorithms and software for the solution of large-scale optimization problems on high-performance architectures. Areas of interest include unconstrained and bound-constrained optimization, nonlinear least squares problems, optimization problems with partial differential equation constraints, and variational inequalities and complementarity constraints. The development of TAO was motivated by the scattered support for parallel computations and the lack of reuse of external toolkits in current optimization software. Our aim is to produce high-quality optimization software for computing environments ranging from workstations and laptops to massively parallel high-performance architectures. Our design decisions are strongly motivated by the challenges inherent in the use of large-scale distributed memory architectures and the reality of working with large, often poorly structured legacy codes for specific applications.

97 MATHEMATICS AND COMPUTING↗

Non-Intrusive Parallel-in-Time Solvers for Partial Differential Equations (Final Report)

Many time-dependent problems and simulations are often modeled using Partial Differential Equations. Traditional modeling approaches that use sequential time-stepping are reaching a bottleneck in optimizing efficiency. The Center of Applied Science and Computing at Lawrence Livermore National Laboratory extensively works on parallelizing these algorithms to leverage the increasing computational power from the growing number of processors in computer hardware. In particular, they aim to design non-intrusive algorithms that can generalize to a variety of problems and sizes without requiring additional information from or modifications on the original problems. Multigrid Reduction in Time (MGRIT) is a parallel-in-time algorithm that is designed to be non-intrusive. This project focuses on increasing the efficiency of MGRIT by approximating the coarse-grid operator using machine learning approaches as a means to find the most non-intrusive, or general, solution.

97 MATHEMATICS AND COMPUTING↗

Solution of partial differential equations on vector and parallel computers

The present status of numerical methods for partial differential equations on vector and parallel computers was reviewed. The relevant aspects of these computers are discussed and a brief review of their development is included, with particular attention paid to those characteristics that influence algorithm selection. Both direct and iterative methods are given for elliptic equations as well as explicit and implicit methods for initial boundary value problems. The intent is to point out attractive methods as well as areas where this class of computer architecture cannot be fully utilized because of either hardware restrictions or the lack of adequate algorithms. Application areas utilizing these computers are briefly discussed.

Ortega, J. M.↗

Solution of partial differential equations on vector and parallel computers

The present status of numerical methods for partial differential equations on vector and parallel computers was reviewed. The relevant aspects of these computers are discussed and a brief review of their development is included, with particular attention paid to those characteristics that influence algorithm selection. Both direct and iterative methods are given for elliptic equations as well as explicit and implicit methods for initial boundary value problems. The intent is to point out attractive methods as well as areas where this class of computer architecture cannot be fully utilized because of either hardware restrictions or the lack of adequate algorithms. Application areas utilizing these computers are briefly discussed.

Ortega, J. M.↗