• Corpus ID: 3334770

Fixed-Rank Approximation of a Positive-Semidefinite Matrix from Streaming Data

  title={Fixed-Rank Approximation of a Positive-Semidefinite Matrix from Streaming Data},
  author={Joel A. Tropp and Alp Yurtsever and Madeleine Udell and Volkan Cevher},
Several important applications, such as streaming PCA and semidefinite programming, involve a large-scale positive-semidefinite (psd) matrix that is presented as a sequence of linear updates. Because of storage limitations, it may only be possible to retain a sketch of the psd matrix. This paper develops a new algorithm for fixed-rank psd approximation from a sketch. The approach combines the Nystrom approximation with a novel mechanism for rank truncation. Theoretical analysis establishes that… 

Figures from this paper

Randomized low-rank approximation for symmetric indefinite matrices

This work identifies the main challenges in developing a Nystr¨om approximation to symmetric indefinite matrices, and establishes relative-error nuclear norm bounds of the resulting approximation that hold when the singular values decay rapidly.

Fast and stable randomized low-rank matrix approximation

This work studies a generalization of Nystr{o}m method applicable to general matrices, and shows that it has near-optimal approximation quality comparable to competing methods and can significantly outperform state-of-the-art methods.

Randomized low-rank approximation of monotone matrix functions

This work presents and analyzes funNystr¨om, a simple and inexpensive method that constructs a low-rank approximation of f ( A ) directly from a NyStr¨om approximation of A, completely bypassing the need for matrix-vector products with f (A ).

Streaming Low-Rank Matrix Approximation with an Application to Scientific Simulation

It is argued that randomized linear sketching is a natural tool for on-the-fly compression of data matrices that arise from large-scale scientific simulations and data collection and is less sensitive to parameter choices than previous techniques.

NysADMM: faster composite convex optimization via low-rank approximation

The breadth of problems on which NysADMM beats standard solvers is a surprise and suggests that ADMM is a dominant paradigm for numerical optimization across a wide range of statistical learning problems that are usually solved with bespoke methods.

Randomized Nystr\"om Preconditioning

Numerical tests show that Nyström PCG can rapidly solve large linear systems that arise in data analysis problems, and it surpasses several competing methods from the literature.

Randomly pivoted Cholesky: Practical approximation of a kernel matrix with few entry evaluations

Empirically, RPC HOLESKY matches or improves on the performance of alternative algorithms for low-rank psd approximation and provably achieves near-optimal approximation guarantees.

Randomized Clustered Nystrom for Large-Scale Kernel Machines

This paper introduces a randomized algorithm for generating landmark points that is scalable to large high-dimensional data sets and performs K-means clustering on low-dimensional random projections of a data set and thus leads to significant savings for high- dimensional data sets.

Structured Stochastic Quasi-Newton Methods for Large-Scale Optimization Problems

Numerical experiments on logistic regression, deep autoencoder networks and deep learning problems show that the efficiency of the proposed method is at least comparable with the state-of-the-art methods.



Randomized single-view algorithms for low-rank matrix approximation

A suite of algorithms for constructing low-rank approximations of an input matrix from a random linear image of the matrix, called a sketch, that can preserve structural properties of the input matrix, such as positive-semidefiniteness, and produce approximation with a user-specified rank are developed.

Numerical linear algebra in the streaming model

Near-optimal space bounds are given in the streaming model for linear algebra problems that include estimation of matrix products, linear regression, low-rank approximation, and approximation of matrix rank; results for turnstile updates are proved.

Revisiting the Nystrom Method for Improved Large-scale Machine Learning

An empirical evaluation of the performance quality and running time of sampling and projection methods on a diverse suite of SPSD matrices and a suite of worst-case theoretical bounds for both random sampling and random projection methods are complemented.

Subspace Iteration Randomization and Singular Value Problems

  • M. Gu
  • Computer Science
    SIAM J. Sci. Comput.
  • 2015
A novel error analysis is presented that considers randomized algorithms within the subspace iteration framework and shows with very high probability that highly accurate low-rank approximations as well as singular values can be computed.

Sketchy Decisions: Convex Low-Rank Matrix Optimization with Optimal Storage

This paper proposes the first algorithm to offer provable convergence to an optimal point with an optimal memory footprint, and modifies a standard convex optimization method to work on a sketched version of the decision variable, and can recover the solution from this sketch.

Topics in randomized numerical linear algebra

The matrix Laplace transform framework is extended to derive Chernoff and Bernstein inequalities that apply to all the eigenvalues of certain classes of random matrices, and to derive convergence rates for each individual eigenvalue of a sample covariance matrix.

Dimensionality Reduction for k-Means Clustering and Low Rank Approximation

This work shows how to approximate a data matrix A with a much smaller sketch ~A that can be used to solve a general class of constrained k-rank approximation problems to within (1+ε) error, and gives a simple alternative to known algorithms that has applications in the streaming setting.

Low-Rank PSD Approximation in Input-Sparsity Time

This work gives algorithms for approximation by low-rank positive semidefinite (PSD) matrices, and shows that there are asymmetric input matrices that cannot have good symmetric column-selected approximations.

Finding Structure with Randomness: Probabilistic Algorithms for Constructing Approximate Matrix Decompositions

This work surveys and extends recent research which demonstrates that randomization offers a powerful tool for performing low-rank matrix approximation, and presents a modular framework for constructing randomized algorithms that compute partial matrix decompositions.

Randomized Clustered Nystrom for Large-Scale Kernel Machines

This paper introduces a randomized algorithm for generating landmark points that is scalable to large high-dimensional data sets and performs K-means clustering on low-dimensional random projections of a data set and thus leads to significant savings for high- dimensional data sets.