Search NASA⌕ Search

SEARCH · Search NASA

Results for “stochastic gradient methods”

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 181 records · Page 10

Extended geometric scaling from generalized traveling waves

We define a mapping of the QCD Balitsky-Kovchegov equation in the diffusive approximation with noise and a generalized coupling allowing a common treatment of the fixed and running QCD couplings. It corresponds to the extension of the stochastic Fisher and Kolmogorov-Petrovskii-Piscounov equation to the radial wave propagation in a medium with negative-gradient absorption responsible for anomalous diffusion, noninteger dimension, and damped noise fluctuations. We obtain its analytic traveling-wave solutions with a new scaling curve and for running coupling a new scaling variable allowing to extend the range and validity of the geometric-scaling QCD prediction beyond the previously known domain.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Output Feedback Stabilization for a Class of Multi-Variable Bilinear Stochastic Systems with Stochastic Coupling Attenuation

In this paper, stochastic coupling attenuation is investigated for a class of multi-variable bilinear stochastic systems and a novel output feedback m-block backstepping controller with linear estimator is designed, where gradient descent optimization is used to tune the design parameters of the controller. It has been shown that the trajectories of the closed-loop stochastic systems are bounded in probability sense and the stochastic coupling of the system outputs can be effectively attenuated by the proposed control algorithm. Moreover, the stability of the stochastic systems is analyzed and the effectiveness of the proposed method has been demonstrated using a simulated example.

47 OTHER INSTRUMENTATION↗

SYSTEMATIC AND STOCHASTIC VARIATIONS IN PULSAR DISPERSION MEASURES

We analyze deterministic and random temporal variations in the dispersion measure (DM) from the full three-dimensional velocities of pulsars with respect to the solar system, combined with electron-density variations over a wide range of length scales. Previous treatments have largely ignored pulsars’ changing distances while favoring interpretations involving changes in sky position from transverse motion. Linear trends in pulsar DMs observed over 5–10 year timescales may signify sizable DM gradients in the interstellar medium (ISM) sampled by the changing direction of the line of sight to the pulsar. We show that motions parallel to the line of sight can also account for linear trends, for the apparent excess of DM variance over that extrapolated from scintillation measurements, and for the apparent non-Kolmogorov scalings of DM structure functions inferred in some cases. Pulsar motions through atomic gas may produce bow-shock ionized gas that also contributes to DM variations. We discuss the possible causes of periodic or quasi-periodic changes in DM, including seasonal changes in the ionosphere, annual variations of the solar elongation angle, structure in the heliosphere and ISM boundary, and substructure in the ISM. We assess the solar cycle’s role on the amplitude of ionospheric and solar wind variations. Interstellar refraction can produce cyclic timing variations from the error in transforming arrival times to the solar system barycenter. We apply our methods to DM time series and DM gradient measurements in the literature and assess their consistency with a Kolmogorov medium. Finally, we discuss the implications of DM modeling in precision pulsar timing experiments.

79 ASTRONOMY AND ASTROPHYSICS↗

Non-adiabatic dynamics around a conical intersection with surface-hopping coupled coherent states

A surface-hopping extension of the coupled coherent states-method [D. Shalashilin and M. Child, Chem. Phys. 304, 103-120 (2004)] for simulating non-adiabatic dynamics with quantum effects of the nuclei is put forward. The time-dependent Schrödinger equation for the motion of the nuclei is solved in a moving basis set. The basis set is guided by classical trajectories, which can hop stochastically between different electronic potential energy surfaces. The non-adiabatic transitions are modelled by a modified version of Tully’s fewest switches algorithm. The trajectories consist of Gaussians in the phase space of the nuclei (coherent states) combined with amplitudes for an electronic wave function. The time-dependent matrix elements between different coherent states determine the amplitude of each trajectory in the total multistate wave function; the diagonal matrix elements determine the hopping probabilities and gradients. In this way, both interference effects and non-adiabatic transitions can be described in a very compact fashion, leading to the exact solution if convergence with respect to the number of trajectories is achieved and the potential energy surfaces are known globally. The method is tested on a 2D model for a conical intersection [A. Ferretti, J. Chem. Phys. 104, 5517 (1996)], where a nuclear wavepacket encircles the point of degeneracy between two potential energy surfaces and interferes with itself. These interference effects are absent in classical trajectory-based molecular dynamics but can be fully incorpo rated if trajectories are replaced by surface hopping coupled coherent states.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Simulation of underground gravity gradients from stochastic seismic fields

We present results obtained from a finite-element simulation of seismic displacement fields and of gravity gradients generated by those fields. The displacement field is constructed by a plane-wave model with a 3D isotropic stochastic field and a 2D fundamental Rayleigh field. The plane-wave model provides an accurate representation of stationary fields from distant sources. Underground gravity gradients are calculated as the acceleration of a free test mass inside a cavity. The results are discussed in the context of gravity-gradient noise subtraction in third generation gravitational-wave detectors. Error analysis with respect to the density of the simulated grid leads to a derivation of an improved seismometer placement inside a 3D array which would be used in practice to monitor the seismic field.

99 GENERAL AND MISCELLANEOUS↗

Integral one-point statistical characteristics of vector fields in stochastic magnetohydrodynamic flows

We study integral statistical characteristics of a vector passive tracer (homogeneous at the initial time) in a velocity field that is assumed to be a Gaussian random field homogeneous in space and delta-correlated in time. Such statistical characteristics describe the dynamical system as a whole in the entire space, separating out the field generation processes, which allows us to not digress into details of the dynamics related to the advection of these quantities. The density field gradient (in the general case of a compressible fluid) and the magnetic field vector with its spatial derivatives (in an incompressible fluid) are such a tracer. We study the isotropization in time, helicity, and dissipation of these fields in the absence of molecular diffusion effects. We formulate a method of successive approximations for the variance of the density field and the mean magnetic field energy that allows the solutions valid in the entire time interval to be obtained in the first order in molecular diffusion coefficients.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

A stochastic approach to estimate the uncertainty of dose mapping caused by uncertainties in b-spline registration

Purpose: In fractionated radiation therapy, image guidance with daily tomographic imaging becomes more and more clinical routine. In principle, this allows for daily computation of the delivered dose and for accumulation of these daily dose distributions to determine the actually delivered total dose to the patient. However, uncertainties in the mapping of the images can translate into errors of the accumulated total dose, depending on the dose gradient. In this work, an approach to estimate the uncertainty of mapping between medical images is proposed that identifies areas bearing a significant risk of inaccurate dose accumulation. Methods: This method accounts for the geometric uncertainty of image registration and the heterogeneity of the dose distribution, which is to be mapped. Its performance is demonstrated in context of dose mapping based on b-spline registration. It is based on evaluation of the sensitivity of dose mapping to variations of the b-spline coefficients combined with evaluation of the sensitivity of the registration metric with respect to the variations of the coefficients. It was evaluated based on patient data that was deformed based on a breathing model, where the ground truth of the deformation, and hence the actual true dose mapping error, is known. Results: The proposed approach has the potential to distinguish areas of the image where dose mapping is likely to be accurate from other areas of the same image, where a larger uncertainty must be expected. Conclusions: An approach to identify areas where dose mapping is likely to be inaccurate was developed and implemented. This method was tested for dose mapping, but it may be applied in context of other mapping tasks as well.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Projection-operator methods for classical transport in magnetized plasmas. Part 2. Nonlinear response and the Burnett equations [Projection-operator methods for classical transport in magnetized plasmas. Part 1. Nonlinear response and the Burnett equations]

The time-independent projection-operator formalism for the derivation of Burnett equations is extended and considered in the context of multispecies and magnetized plasmas. The procedure provides specific formulas for transport coefficients in terms of two-time correlation functions involving both two and three phase-space points. It is shown how to calculate those correlation functions in the limit of weak coupling. The results are used to demonstrate, with the aid of a particular non-trivial example, that the Chapman–Enskog methodology employed by Catto & Simakov (CS) (Phys. Plasmas, vol. 11, 2004, pp. 90–102) to calculate the contributions to the parallel viscosity driven by temperature gradients is consistent with formulas previously derived from the two-time formalism by Brey (J. Chem. Phys., vol. 79, 1983, pp. 4585–4598). Here, the work serves to unify previous work on plasma kinetic theory with formalism usually applied to turbulence. Additional contributions include discussions of (i) Braginskii-order interspecies momentum exchange from the point of view of two-time correlations; and (ii) a simple stochastic model, unrelated to many-body theory, that exhibits Burnett effects. Insights from that model emphasize the role of non-Gaussian statistics in the evaluation of Burnett transport coefficients, including the effects calculated by CS that stem from the nonlinear collision operator. Together, Parts 1 and 2 of this series provide an introduction to projection-operator methods that should be broadly useful in theoretical plasma physics.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

A Hybrid Gradient Method to Designing Bayesian Experiments for Implicit Models

Bayesian experimental design (BED) aims at designing an experiment to maximize the information gathering from the collected data. The optimal design is usually achieved by maximizing the mutual information (MI) between the data and the model parameters. When the analytical expression of the MI is unavailable, e.g.,having implicit models with intractable data distributions, a neural network-based lower bound of the MI was recently proposed and a gradient ascent method was used to maximize the lower bound [1]. However, the approach in [1] requires a pathwise sampling path to compute the gradient of the MI lower bound with respect to the design variables, and such a pathwise sampling path is usually inaccessible for implicit models. In this work, we propose a hybrid gradient approach that leverages recent advances in variational MI estimator and evolution strategies (ES)combined with black-box stochastic gradient ascent (SGA) to maximize the MI lower bound. This allows the design process to be achieved through a unified scalable procedure for implicit models without sampling path gradients. Several experiments demonstrate that our approach significantly improves the scalability of BED for implicit models in high-dimensional design space.

Zhang, Jiaxin↗

Concurrent multi-peak Bragg coherent x-ray diffraction imaging of 3D nanocrystal lattice displacement via global optimization

Abstract In this paper we demonstrated a method to reconstruct vector-valued lattice distortion fields within nanoscale crystals by optimization of a forward model of multi-reflection Bragg coherent diffraction imaging (MR-BCDI) data. The method flexibly accounts for geometric factors that arise when making BCDI measurements, is amenable to efficient inversion with modern optimization toolkits, and allows for globally constraining a single image reconstruction to multiple Bragg peak measurements. This is enabled by a forward model that emulates the multiple Bragg peaks of a MR-BCDI experiment from a single estimate of the 3D crystal sample. We present this forward model, we implement it within the stochastic gradient descent optimization framework, and we demonstrate it with simulated and experimental data of nanocrystals with inhomogeneous internal lattice displacement. We find that utilizing a global optimization approach to MR-BCDI affords a reliable path to convergence of data which is otherwise challenging to reconstruct.

36 MATERIALS SCIENCE↗

A Predictive Prescription Framework for Stochastic Unit Commitment Using Boosting Ensemble Learning Algorithms

To take unit commitment (UC) decisions under uncertain load, most existing stochastic optimization (SO) frameworks adopt a generic representation of uncertainty. While load levels that materialize on a particular day are influenced by various covariates (such as the day of the week or temperature), SO frameworks typically disregard such side observations, wasting actionable information that could significantly enhance decision quality. Here, this article proposes a contextual SO (CSO) framework for UC under uncertain load, which can effectively exploit covariate observations in conjunction with a class of machine learning (ML) algorithms to improve the out-of-sample performance of UC decisions. It shows how three ML algorithms, adaptive boosting, gradient boosted trees, and extreme gradient boosting, can be used to this end, constituting the first application of these algorithms in any CSO framework. Using real-world data harvested from the New York ISO grid, we measure the out-of-sample performance of the framework in terms of total operation cost, shed load values, locational marginal prices, and total payments by the loads, against several benchmark methods proposed in the literature. The article has an online companion (Yurdakul et al.), wherein we present additional results and lay out further mathematical formulations used in this work.

42 ENGINEERING↗

Revisiting the ODE Method for Recursive Algorithms: Fast Convergence Using Quasi Stochastic Approximation

Several decades ago, Profs. Sean Meyn and Lei Guo were postdoctoral fellows at ANU, where they shared interest in recursive algorithms. It seems fitting to celebrate Lei Guo's 60th birthday with a review of the ODE Method and its recent evolution. The method has been regarded as a technique for algorithm analysis. It is argued that this viewpoint is backwards: The original stochastic approximation method was surely motivated by an ODE, and tools for analysis came much later (based on establishing robustness of Euler approximations). The paper presents a brief survey of recent research in machine learning that shows the power of algorithm design in continuous time, following by careful approximation to obtain a practical recursive algorithm. While these methods are usually presented in a stochastic setting, this is not a prerequisite. In fact, recent theory shows that rates of convergence can be dramatically accelerated by applying techniques inspired by quasi Monte-Carlo. Subject to conditions, the optimal rate of convergence can be obtained by applying the averaging technique of Polyak and Ruppert. The conditions are not universal, but theory suggests alternatives to achieve acceleration. The theory is illustrated with applications to gradient-free optimization, and policy gradient algorithms for reinforcement learning.

learning and adaptive systems in artificial intell↗

Volt-VAR Optimization in Distribution Networks Using Twin Delayed Deep Reinforcement Learning

Modern distribution grids are undergoing new challenges due to the stochastic nature of distributed energy resources (DERs). High penetration of DERs has a significant impact on Volt-VAR profile and system power losses. This work proposes a deep reinforcement learning (DRL)-based Volt-VAR optimization approach for improving voltage profile and reducing system power loss under high penetration of distributed energy resources, such as battery energy storage and solar photovoltaic units in distribution grids. The twin delayed deep deterministic policy gradient (TD3) method-based DRL agent is proposed to configure optimal set-points of reactive power outputs of fast responding smart inverters. The agent schedules the reactive power of inverters according to their physical capabilities, such as minimum allowed power factor, e.g., 0.9 leading/lagging. The reward function of the proposed DRL scheme is designed carefully to ensure a proper voltage profile of the grids with effective scheduling of reactive power outputs from inverters. The performance of the proposed model is verified on modified IEEE 34- and 123-bus systems and compared with base case with no reactive supply by inverters, and local droop Volt-VAR control approach. The results show that the proposed method performs better than the local droop control and deep deterministic policy gradient (DDPG)-based DRL method for reducing voltage fluctuation and minimizing power loss.

Hossain, Rakib↗

Staggered scheme for the compressible fluctuating hydrodynamics of multispecies fluid mixtures

Here, we present a numerical formulation for the solution of nonisothermal, compressible Navier-Stokes equations with thermal fluctuations to describe mesoscale transport phenomena in multispecies fluid mixtures. The novelty of our numerical method is the use of staggered grid momenta along with a finite volume discretization of the thermodynamic variables to solve the resulting stochastic partial differential equations. The key advantages of the numerical scheme are that it significantly simplifies the discretization of diffusive and stochastic momentum fluxes into a more compact form, and it provides an unambiguous prescription of boundary conditions involving pressure. The staggered grid scheme more accurately reproduces the equilibrium static structure factor of hydrodynamic fluctuations in gas mixtures compared to a collocated scheme described previously by Balakrishnan et al. [Phys. Rev. E 89, 013017 (2014)1539-375510.1103/PhysRevE.89.013017]. The numerical method is tested for ideal noble gases mixtures under various nonequilibrium conditions, such as applied thermal and concentration gradients, to assess the role of cross-diffusion effects, such as Soret and Dufour, on the long-ranged correlations of hydrodynamic fluctuations, which are also more accurately reproduced compared to the collocated scheme. We numerically study giant nonequilibrium fluctuations driven by concentration gradients and fluctuation-driven Rayleigh-Taylor instability in gas mixtures. Wherever applicable, excellent agreement is observed with theory and measurements from the direct simulation Monte Carlo method.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Trustworthy Physics-Informed Deep Learning for Predictive Scientific Computing

This project has developed powerful trustworthy physics-informed deep learning (TPiDL) models and methods to fundamentally enhance the scale and power of computational modeling in the scientific and engineering domains. Deep learning (DL) has radically advanced the state-of-the-art in machine learning, computer vision, natural language processing, and also scientific computing. Nevertheless, progress has been driven almost entirely by empirical observations, hacks, and tricks. Under the support of this project, the graph operator learning tools and advanced trustworthy physical informed neural networks have been developed. In addition, stochastic gradient replica-exchange Markov Chain Monte Carlo (MCMC) sampling algorithms have been designed to quantify the uncertainties and speed up the training of large-scale neural networks.

97 MATHEMATICS AND COMPUTING↗

Stochastic inversion of permeability and dispersivities from time lapse self‐potential measurements: A controlled sandbox study

We test the ability of the self‐potential method to provide information that can be used to invert the permeability and the dispersivities of a porous material. We first formulate the semi‐coupled system of equations describing the occurrence of self‐potential signals associated with the transport of salt in a porous material like sand. Two contributions in the source current density arise from the electrokinetic coupling associated with the flow of the pore water and the diffusion potential due to the gradient in the activity of the salt. A controlled laboratory sandbox experiment is performed to show that time‐lapse self‐potential measurements can be used to invert, in a Bayesian framework, the permeability and dispersivities.

self-potential, dispersivity, permeability↗

Correspondence between neuroevolution and gradient descent

Abstract We show analytically that training a neural network by conditioned stochastic mutation or neuroevolution of its weights is equivalent, in the limit of small mutations, to gradient descent on the loss function in the presence of Gaussian white noise. Averaged over independent realizations of the learning process, neuroevolution is equivalent to gradient descent on the loss function. We use numerical simulation to show that this correspondence can be observed for finite mutations, for shallow and deep neural networks. Our results provide a connection between two families of neural-network training methods that are usually considered to be fundamentally different.

97 MATHEMATICS AND COMPUTING↗