Search NASA⌕ Search

SEARCH · Search NASA

Results for “subspace method”

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 199 records · Page 11

Nuclear-norm-based Subspace Identification of Multi-Zone Building HVAC System

There have been many attempts to reduce the uncertainty associated with renewable energy. This paper aims to contribute to the field by considering the building heating, ventilation and air conditioning (HVAC) system, which is treated as flexible load. The control of the indoor temperature in the building employing the locally generated renewable energy has been carried out with both model-based and data-driven approaches. Although numerous efforts have been made in the area of construction of a model for the building HVAC system, almost all models have shortcomings. A building model that is locally accurate is required to be able to predict the load consumption in the future. This will benefit both the utility companies and building managers to estimate the upcoming load shape, especially the peak amount and duration. We consider the case of a multi-zone building HVAC system, i.e., each zone has its own thermostat and can be set to a different temperature. Based on our need for a multi-zone model that is locally accurate, we resort to the method of system identification. In particular, we will employ the nuclear norm based subspace identification method.

Telsang, Bhagyashri↗

A new classification scheme of plastic wastes based upon recycling labels

Highlights: • PET, HPDE or PP types of plastics are considered. • An automated classification of plastic bottles based on the feature extraction and classification methods is performed. • The decision mechanism consists of PCA, Kernel PCA, FLDA, SVD and Laplacian Eigenmaps methods. • SVM is selected to achieve the classification task and majority voting technique is used. - Abstract: Since recycling of materials is widely assumed to be environmentally and economically beneficial, reliable sorting and processing of waste packaging materials such as plastics is very important for recycling with high efficiency. An automated system that can quickly categorize these materials is certainly needed for obtaining maximum classification while maintaining high throughput. In this paper, first of all, the photographs of the plastic bottles have been taken and several preprocessing steps were carried out. The first preprocessing step is to extract the plastic area of a bottle from the background. Then, the morphological image operations are implemented. These operations are edge detection, noise removal, hole removing, image enhancement, and image segmentation. These morphological operations can be generally defined in terms of the combinations of erosion and dilation. The effect of bottle color as well as label are eliminated using these operations. Secondly, the pixel-wise intensity values of the plastic bottle images have been used together with the most popular subspace and statistical feature extraction methods to construct the feature vectors in this study. Only three types of plastics are considered due to higher existence ratio of them than the other plastic types in the world. The decision mechanism consists of five different feature extraction methods including as Principal Component Analysis (PCA), Kernel PCA (KPCA), Fisher’s Linear Discriminant Analysis (FLDA), Singular Value Decomposition (SVD) and Laplacian Eigenmaps (LEMAP) and uses a simple experimental setup with a camera and homogenous backlighting. Due to the giving global solution for a classification problem, Support Vector Machine (SVM) is selected to achieve the classification task and majority voting technique is used as the decision mechanism. This technique equally weights each classification result and assigns the given plastic object to the class that the most classification results agree on. The proposed classification scheme provides high accuracy rate, and also it is able to run in real-time applications. It can automatically classify the plastic bottle types with approximately 90% recognition accuracy. Besides this, the proposed methodology yields approximately 96% classification rate for the separation of PET or non-PET plastic types. It also gives 92% accuracy for the categorization of non-PET plastic types into HPDE or PP.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Dynamical eightfold way mesons in strongly coupled lattice QCD

We consider a 3+1 lattice QCD model with three quark flavors, local SU(3){sub c} gauge symmetry, global SU(3){sub f} flavor symmetry, in an imaginary-time formulation and with strong coupling (a small hopping parameter {kappa}>0 and a plaquette coupling {beta}>0, 0<{beta}<<{kappa}<<1). Associated with the model there is an underlying physical quantum mechanical Hilbert space H which, via a Feynman-Kac formula, enables us to derive spectral representations for correlations and obtain the low-lying energy-momentum spectrum exactly. Using the decoupling of hyperplane method and concentrating on the subspace H{sub e} subset of H of vectors with an even number of quarks, we obtain the one-particle spectrum showing the existence of 36 meson states from dynamical first principles, i.e. directly from the quark-gluon dynamics. The particles are detected by isolated dispersion curves w(p-vector) in the energy-momentum spectrum. Besides the SU(3){sub f} quantum numbers (total hypercharge, quadratic Casimir C{sub 2}, total isospin and its 3rd component), the basic excitations also carry spin labels. The total spin operator J and its z-component J{sub z} are defined using {pi}/2 rotations about the spatial coordinate axes and agree with the infinitesimal generators of the continuum for improper zero-momentum meson states. The eightfold way meson particles are given by linear combinations of these 36 states and can be grouped into three SU(3){sub f} nonets associated with the vector mesons (J=1, J{sub z}=0, {+-}1) and one nonet associated with the pseudoscalar mesons (J=0). Each nonet admits a further decomposition into a SU(3){sub f} singlet (C{sub 2}=0) and an octet (C{sub 2}=3). For {beta}=0, the particle dispersion curves are all of the form w(p-vector)=-2ln{kappa}-3{kappa}{sup 2}/2+(1/4){kappa}{sup 2}{sub j=1}{sup 3}2(1-cosp{sup j})+{kappa}{sup 4}r({kappa},p-vector), with p-vector(set-membership sign)(-{pi},{pi}]{sup 3} and |r({kappa},p-vector)|{<=}const. For the pseudoscalar mesons, r({kappa},p-vector) is jointly analytic in {kappa} and p{sup j}, for |{kappa}| and |Imp{sup j}| small. At {beta}=0 the meson masses are given by m({kappa})=-2ln{kappa}-3{kappa}{sup 2}/2+{kappa}{sup 4}r({kappa}), with r(0){ne}0 and r({kappa}) real analytic; for {beta}{ne}0 the nonsingular part of the mass, m({kappa},{beta})+2ln{kappa}, is jointly analytic in {kappa} and {beta}. For a fixed nonet, the mass of the vector mesons are independent of J{sub z} and are all equal within each octet. All singlet masses are also equal for the vector mesons. For {beta}=0, up to and including O({kappa}{sup 4}), for each nonet, the masses of the octet and the singlet are found to be equal. All members of each octet have identical dispersions. Other dispersion curves may differ. Indeed, there is a pseudoscalar, vector meson mass splitting (between J=0 and J=1) given by 2{kappa}{sup 4}+O({kappa}{sup 6}) at {beta}=0, analytic in {beta} and the splitting persists for {beta}<<{kappa}. Using a correlation subtraction method, we show the 36 meson states give the only spectrum in H{sub e} up to near the two-meson threshold of {approx_equal}-4ln{kappa}. Combining our present result with a similar one for baryons (of asymptotic mass -3ln{kappa}) shows that the model does exhibit confinement up to near the two-meson threshold.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Learning physics-based reduced-order models from data using nonlinear manifolds

Here we present a novel method for learning reduced-order models of dynamical systems using nonlinear manifolds. First, we learn the manifold by identifying nonlinear structure in the data through a general representation learning problem. The proposed approach is driven by embeddings of low-order polynomial form. A projection onto the nonlinear manifold reveals the algebraic structure of the reduced-space system that governs the problem of interest. The matrix operators of the reduced-order model are then inferred from the data using operator inference. Numerical experiments on a number of nonlinear problems demonstrate the generalizability of the methodology and the increase in accuracy that can be obtained over reduced-order modeling methods that employ a linear subspace approximation.

97 MATHEMATICS AND COMPUTING↗

Quantum Filtering and Analysis of Multiplicities in Eigenvalue Spectra

Fine-grained spectral properties of quantum Hamiltonians, including both eigenvalues and their multiplicities, provide useful information for characterizing many-body quantum systems as well as for understanding phenomena such as topological order. Extracting such information with small additive error is #BQP-complete in the worst case. In this work, we introduce QFAMES (quantum filtering and analysis of multiplicities in eigenvalue spectra), a quantum algorithm that efficiently identifies clusters of closely spaced dominant eigenvalues and determines their multiplicities under physically motivated assumptions, which allows us to bypass worst-case complexity barriers. QFAMES also enables the estimation of observable expectation values within targeted energy clusters, providing a powerful tool for studying quantum phase transitions and other physical properties. We validate the effectiveness of QFAMES through numerical demonstrations, including its applications to characterizing quantum phases in the transverse-field Ising model and estimating the ground-state degeneracy of a topologically ordered phase in the two-dimensional toric code model. We also generalize QFAMES to the setting of mixed initial states. Our approach offers rigorous theoretical guarantees and significant advantages over existing subspace-based quantum spectral analysis methods, particularly in terms of the sample complexity and the ability to resolve degeneracies.

97 MATHEMATICS AND COMPUTING↗

Decoherence-free neutron interferometry

Perfect single-crystal neutron interferometers are adversely sensitive to environmental disturbances, particularly mechanical vibrations. The sensitivity to vibrations results from the slow velocity of thermal neutrons and the long measurement time that are encountered in a typical experiment. Consequently, to achieve a good interference solutions for reducing vibration other than those normally used in optical experiments must be explored. Here we introduce a geometry for a neutron interferometer that is less sensitive to low-frequency vibrations. This design may be compared with both dynamical decoupling methods and decoherence-free subspaces that are described in quantum information processing. By removing the need for bulky vibration isolation setups, this design will make it easier to adopt neutron interferometry to a wide range of applications and increase its sensitivity.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Emulating ab initio computations of infinite nucleonic matter

We construct efficient emulators for the computation of the infinite nuclear matter equation of state. These emulators are based on the subspace-projected coupled-cluster method for which we here develop a new algorithm called small-batch voting to eliminate spurious states that might appear when emulating quantum many-body methods based on a non-Hermitian Hamiltonian. The efficiency and accuracy of these emulators facilitate a rigorous statistical analysis within which we explore nuclear matter predictions for > 10 6 different parametrizations of a chiral interaction model with explicit Δ -isobars at next-to-next-to leading order. Constrained by nucleon-nucleon scattering phase shifts and bound-state observables of light nuclei up to He 4 , we use history matching to identify nonimplausible domains for the low-energy coupling constants of the chiral interaction. Within these domains we perform a Bayesian analysis using sampling and importance resampling with different likelihood calibrations and study correlations between interaction parameters, calibration observables in light nuclei, and nuclear matter saturation properties. Published by the American Physical Society 2024

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Space&ndash;Time Least-Squares Petrov&ndash;Galerkin Projection for Nonlinear Model Reduction

This work proposes a space--time least-squares Petrov--Galerkin (ST-LSPG) projection method for model reduction of nonlinear dynamical systems. In contrast to typical nonlinear model-reduction methods that first apply (Petrov--)Galerkin projection in the spatial dimension and subsequently apply time integration to numerically resolve the resulting low-dimensional dynamical system, the proposed method applies projection in space and time simultaneously. To accomplish this, the method first introduces a low-dimensional space--time trial subspace, which can be obtained by computing tensor decompositions of state-snapshot data. The method then computes discrete-optimal approximations in this space--time trial subspace by minimizing the residual arising after time discretization over all space and time in a weighted $\ell^2$-norm. This norm can be defined to enable complexity reduction (i.e., hyper-reduction) in time, which leads to space--time collocation and space--time Gauss--Newton with Approximated Tensors (GNAT) variants of the ST-LSPG method. Advantages of the approach relative to typical spatial-projection-based nonlinear model reduction methods such as Galerkin projection and least-squares Petrov--Galerkin projection include a reduction of both the spatial and temporal dimensions of the dynamical system, and a priori error bounds that bound the solution error by the best space--time approximation error and whose stability constants exhibit slower growth in time. Numerical examples performed on model problems in fluid dynamics demonstrate the ability of the method to generate orders-of-magnitude computational savings relative to spatial-projection-based reduced-order models without sacrificing accuracy for a fixed spatio-temporal discretization.

97 MATHEMATICS AND COMPUTING↗

Stochastic Least-Squares Petrov--Galerkin Method for Parameterized Linear Systems

We consider the numerical solution of parameterized linear systems where the system matrix, the solution, and the right-hand side are parameterized by a set of uncertain input parameters. We explore spectral methods in which the solutions are approximated in a chosen finite-dimensional subspace. It has been shown that the stochastic Galerkin projection technique fails to minimize any measure of the solution error [A. Mugler and H.-J. Starkloff, ESAIM Math. Model. Numer. Anal., 47 (2013), pp. 1237--1263]. As a remedy for this, we propose a novel stochatic least-squares Petrov--Galerkin (LSPG) method. The proposed method is optimal in the sense that it produces the solution that minimizes a weighted $\ell^2$-norm of the residual over all solutions in a given finite-dimensional subspace. Moreover, the method can be adapted to minimize the solution error in different weighted $\ell^2$-norms by simply applying a weighting function within the least-squares formulation. In addition, a goal-oriented seminorm induced by an output quantity of interest can be minimized by defining a weighting function as a linear functional of the solution. We establish optimality and error bounds for the proposed method, and extensive numerical experiments show that the weighted LSPG method outperforms other spectral methods in minimizing corresponding target weighted norms.

97 MATHEMATICS AND COMPUTING↗

Cross-Domain Recognition by Identifying Joint Subspaces of Source Domain and Target Domain

This paper introduces here a new method to solve the cross-domain recognition problem. Different from the traditional domain adaption methods which rely on a global domain shift for all classes between the source and target domains, the proposed method is more flexible to capture individual class variations across domains. By adopting a natural and widely used assumption that the data samples from the same class should lay on an intrinsic low-dimensional subspace, even if they come from different domains, the proposed method circumvents the limitation of the global domain shift, and solves the cross-domain recognition by finding the joint subspaces of the source and target domains. Specifically, given labeled samples in the source domain, we construct a subspace for each of the classes. Then we construct subspaces in the target domain, called anchor subspaces, by collecting unlabeled samples that are close to each other and are highly likely to belong to the same class. The corresponding class label is then assigned by minimizing a cost function which reflects the overlap and topological structure consistency between subspaces across the source and target domains, and within the anchor subspaces, respectively. We further combine the anchor subspaces to the corresponding source subspaces to construct the joint subspaces. Subsequently, one-versus-rest support vector machine classifiers are trained using the data samples belonging to the same joint subspaces and applied to unlabeled data in the target domain. We evaluate the proposed method on two widely used datasets: 1) object recognition dataset for computer vision tasks and 2) sentiment classification dataset for natural language processing tasks. Comparison results demonstrate that the proposed method outperforms the comparison methods on both datasets.

97 MATHEMATICS AND COMPUTING↗

Application of an efficient generator-coordinate subspace-selection algorithm to neutrinoless double- β decay

The generator coordinate method begins with the variational construction of a set of nonorthogonal mean-field states that span a subspace of the full many-body Hilbert space. These states are then often projected onto states with good quantum numbers to restore symmetries, leading to a set with members that can be similar to one another, and it is sometimes possible to reduce this set without greatly affecting results. Here, we propose a greedy algorithm that we call the energy-transition-orthogonality procedure (ENTROP) to select subsets of important states. As applied here, the approach selects on the basis of diagonal energy, orthogonality, and contribution to the matrix element that governs neutrinoless double-β decay. We present both shell-model and preliminary ab initio calculations of this matrix element for the decay of 76Ge, with quadrupole deformation parameters and the isoscalar pairing strength as generator coordinates. ENTROP converges quickly, reducing significantly the number of basis states needed for an accurate calculation.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Objected constrained registration and manifold learning: A new patient setup approach in image guided radiation therapy of thoracic cancer

Purpose: The management of thoracic malignancies with radiation therapy is complicated by continuous target motion. In this study, a real time motion analysis approach is proposed to improve the accuracy of patient setup. Methods: For 11 lung cancer patients a long training fluoroscopy was acquired before the first treatment, and multiple short testing fluoroscopies were acquired weekly at the pretreatment patient setup of image guided radiotherapy (IGRT). The data analysis consisted of three steps: first a 4D target motion model was constructed from 4DCT and projected to the training fluoroscopy through deformable registration. Then the manifold learning method was used to construct a 2D subspace based on the target motion (kinetic) and location (static) information in the training fluoroscopy. Thereafter the respiratory phase in the testing fluoroscopy was determined by finding its location in the subspace. Finally, the phase determined testing fluoroscopy was registered to the corresponding 4DCT to derive the pretreatment patient position adjustment for the IGRT. The method was tested on clinical image sets and numerical phantoms. Results: The registration successfully reconstructed the 4D motion model with over 98% volume similarity in 4DCT, and over 95% area similarity in the training fluoroscopy. The machine learning method derived the phase values in over 98% and 93% test images of the phantom and patient images, respectively, with less than 3% phase error. The setup approach achieved an average accumulated setup error less than 1.7 mm in the cranial-caudal direction and less than 1 mm in the transverse plane. All results were validated against the ground truth of manual delineations by an experienced radiation oncologist. The expected total time for the pretreatment setup analysis was less than 10 s. Conclusions: By combining the registration and machine learning, the proposed approach has the potential to improve the accuracy of pretreatment setup for patients with thoracic malignancy.

60 APPLIED LIFE SCIENCES↗

Space-time least-squares Petrov-Galerkin projection in nonlinear model reduction.

Our work proposes a space-time least-squares Petrov-Galerkin (ST-LSPG) projection method for model reduction of nonlinear dynamical systems. In contrast to typical nonlinear model-reduction methods that first apply Petrov-Galerkin projection in the spatial dimension and subsequently apply time integration to numerically resolve the resulting low-dimensional dynamical system, the proposed method applies projection in space and time simultaneously. To accomplish this, the method first introduces a low-dimensional space-time trial subspace, which can be obtained by computing tensor decompositions of state-snapshot data. The method then computes discrete-optimal approximations in this space-time trial subspace by minimizing the residual arising after time discretization over all space and time in a weighted ℓ 2 -norm. This norm can be de ned to enable complexity reduction (i.e., hyper-reduction) in time, which leads to space-time collocation and space-time GNAT variants of the ST-LSPG method. Advantages of the approach relative to typical spatial-projection-based nonlinear model reduction methods such as Galerkin projection and least-squares Petrov-Galerkin projection include: (1) a reduction of both the spatial and temporal dimensions of the dynamical system, (2) the removal of spurious temporal modes (e.g., unstable growth) from the state space, and (3) error bounds that exhibit slower growth in time. Numerical examples performed on model problems in fluid dynamics demonstrate the ability of the method to generate orders-of-magnitude computational savings relative to spatial-projection-based reduced-order models without sacrificing accuracy.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Massively parallel solvers for elliptic partial differential equations in numerical weather and climate prediction

The demand for substantial increases in the spatial resolution of global weather and climate prediction models makes it essential to use numerically efficient and highly scalable algorithms to solve the equations of large-scale atmospheric fluid dynamics. For stability and efficiency reasons, several of the operational forecasting centres, in particular the Met Office and the European Centre for Medium-Range Weather Forecasts (ECMWF) in the UK, use semi-implicit semi-Lagrangian time-stepping in the dynamical core of the model. The additional burden with this method is that a three-dimensional elliptic partial differential equation (PDE) for the pressure correction has to be solved at every model time step and this often constitutes a significant proportion of the time spent in the dynamical core. In global models, this PDE must be solved in a thin spherical shell. To run within tight operational time-scales, the solver has to be parallelized and there seems to be a (perceived) misconception that elliptic solvers do not scale to large processor counts and hence implicit time-stepping cannot be used in very high-resolution global models. After reviewing several methods for solving the elliptic PDE for the pressure correction and their application in atmospheric models, we demonstrate the performance and very good scalability of Krylov subspace solvers and multigrid algorithms for a representative model equation with more than 1010 unknowns on 65 536 cores on the High-End Computing Terascale Resource (HECToR), the UK's national supercomputer. Here, we tested and optimized solvers from two existing numerical libraries (the Distributed and Unified Numerics Environment (DUNE) and Parallel High Performance Preconditioners (hypre)) and implemented both a conjugate gradient solver and a geometric multigrid algorithm based on a tensor-product approach, which exploits the strong vertical anisotropy of the discretized equation. We study both weak and strong scalability and compare the absolute solution times for all methods; in contrast to one-level methods, the multigrid solver is robust with respect to parameter variations.

58 GEOSCIENCES↗

XCOM intrinsic dimensionality for low-Z elements at diagnostic energies

Purpose: To determine the intrinsic dimensionality of linear attenuation coefficients (LACs) from XCOM for elements with low atomic number (Z = 1-20) at diagnostic x-ray energies (25-120 keV). H{sub 0}{sup q}, the hypothesis that the space of LACs is spanned by q bases, is tested for various q-values. Methods: Principal component analysis is first applied and the LACs are projected onto the first q principal component bases. The residuals of the model values vs XCOM data are determined for all energies and atomic numbers. Heteroscedasticity invalidates the prerequisite of i.i.d. errors necessary for bootstrapping residuals. Instead wild bootstrap is applied, which, by not mixing residuals, allows the effect of the non-i.i.d residuals to be reflected in the result. Credible regions for the eigenvalues of the correlation matrix for the bootstrapped LAC data are determined. If subsequent credible regions for the eigenvalues overlap, the corresponding principal component is not considered to represent true data structure but noise. If this happens for eigenvalues l and l + 1, for any l{<=}q, H{sub 0}{sup q} is rejected. Results: The largest value of q for which H{sub 0}{sup q} is nonrejectable at the 5%-level is q = 4. This indicates that the statistically significant intrinsic dimensionality of low-Z XCOM data at diagnostic energies is four. Conclusions: The method presented allows determination of the statistically significant dimensionality of any noisy linear subspace. Knowledge of such significant dimensionality is of interest for any method making assumptions on intrinsic dimensionality and evaluating results on noisy reference data. For LACs, knowledge of the low-Z dimensionality might be relevant when parametrization schemes are tuned to XCOM data. For x-ray imaging techniques based on the basis decomposition method (Alvarez and Macovski, Phys. Med. Biol. 21, 733-744, 1976), an underlying dimensionality of two is commonly assigned to the LAC of human tissue at diagnostic energies. The finding of a higher statistically significant dimensionality thus raises the question whether a higher assumed model dimensionality (now feasible with the advent of multibin x-ray systems) might also be practically relevant, i.e., if better tissue characterization results can be obtained.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Characteristics of a Two-Dimensional Hydrogen-like Atom

Using the customary and well-known representation of the radiation probability of a hydrogen-like atom in the three-dimensional case, a general expression for the probability of single-photon emission of a twodimensional atom has been obtained along with an expression for the particular case of the transition from the first excited state to the ground state, in the latter case in comparison with corresponding expressions for the three-dimensional atom and the one-dimensional atom. Arguments are presented in support of the claim that this method of calculation gives a value of the probability that is identical to the value given by exact methods of QED extended to the subspace {0, 1, 2}. Relativistic corrections ~ (Zα){sup 4} to the usual Schrödinger value of the energy (~ (Zα){sup 2}) are also discussed.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Edgewise Structural Damping of a 2.8-MW Land-Based Wind Turbine Rotor Blade

Modern wind turbines push the predictive capabilities of state-of-the-art aero-servo-elastic tools. The existing limitations hide across the numerical tool chain and can result in serious issues, such as missing aeroelastic instabilities during the design phase. Structural damping is an input that is usually hard to estimate, but also has a major impact on the turbine behavior. In this paper, we discuss an experiment that aims to accurately quantify the structural damping characterizing the edgewise modes of modern wind turbine blades. The experiment is carried out on a 2.8-MW land-based wind turbine and features a fast yaw actuation that induces an edgewise motion on one of the three blades. The Covariant-subspace system identification (Cov-SSI) method is then used to post-process the blade root moment to estimate the short-term edgewise structural damping. Despite limitations of the Cov-SSI method, which consistently under-predicts the absolute values of damping, we observe that structural damping decreases across the first three blade edgewise modes, which is different from the stiffness-proportional damping model that assumes that structural damping increases with the modes. This paper argues that a stiffness-proportional damping model, which is implemented in most aeroelastic tools, is therefore not conservative and might hide aeroelastic instabilities that can instead appear in the field.

17 WIND ENERGY↗

Gradient Coding With Iterative Block Leverage Score Sampling

Gradient coding is a method for mitigating straggling servers in a centralized computing network that uses erasure-coding techniques to distributively carry out first-order optimization methods. Randomized numerical linear algebra uses randomization to develop improved algorithms for large-scale linear algebra computations. In this study, we propose a method for distributed optimization that combines gradient coding and randomized numerical linear algebra. The proposed method uses a randomized ℓ 2 -subspace embedding and a gradient coding technique to distribute blocks of data to the computational nodes of a centralized network, and at each iteration the central server only requires a small number of computations to obtain the steepest descent update. The novelty of our approach is that the data is replicated according to importance scores, called block leverage scores, in contrast to most gradient coding approaches that uniformly replicate the data blocks. Furthermore, we do not require a decoding step at each iteration, avoiding a bottleneck in previous gradient coding schemes. We show that our approach results in a valid ℓ 2 -subspace embedding, and that our resulting approximation converges to the optimal solution.

97 MATHEMATICS AND COMPUTING↗