Low-distortion subspace embeddings in input-sparsity time and applications to robust linear regression

DOI10.1145/2488608.2488621MaRDI QIDQ5495779zbMATH OpenOpenAlexFDO

Authors Xiangrui Meng, Michael W. Mahoney

Publication date 7 August 2014

Published in Proceedings of the forty-eighth annual ACM symposium on Theory of Computing (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1210.3135

zbMATH Keywords

linear regressionrobust regressionPsubspace embeddinglow-distortion embeddinginput-sparsity time

Mathematics Subject Classification ID

Linear regression; mixed models (62J05) Analysis of algorithms and problem complexity (68Q25) Randomized algorithms (68W20) Computational difficulty of problems (lower bounds, completeness, difficulty of approximation, etc.) (68Q17)

Abstract: Low-distortion embeddings are critical building blocks for developing random sampling and random projection algorithms for linear algebra problems. We show that, given a matrix $Ain{R}^{nimesd}$ with $nggd$ and a $pin[1,2)$, with a constant probability, we can construct a low-distortion embedding matrix $Piin{R}^{O(poly(d))imesn}$ that embeds $A_p$, the $el{l}_p$ subspace spanned by $A$'s columns, into $(R^{O(poly(d))},|cdot{|}_p)$; the distortion of our embeddings is only $O(poly(d))$, and we can compute $PiA$ in $O(nz(A))$ time, i.e., input-sparsity time. Our result generalizes the input-sparsity time $el{l}_2$ subspace embedding by Clarkson and Woodruff [STOC'13]; and for completeness, we present a simpler and improved analysis of their construction for $el{l}_2$. These input-sparsity time $el{l}_p$ embeddings are optimal, up to constants, in terms of their running time; and the improved running time propagates to applications such as $(1pmepsilon)$-distortion $el{l}_p$ subspace embedding and relative-error $el{l}_p$ regression. For $el{l}_2$, we show that a $(1+epsilon)$-approximate solution to the $el{l}_2$ regression problem specified by the matrix $A$ and a vector $bin{R}^n$ can be computed in $O(nz(A)+d^{3}log(d/epsilon)/epsilo{n}^2)$ time; and for $el{l}_p$, via a subspace-preserving sampling procedure, we show that a $(1pmepsilon)$-distortion embedding of $A_p$ into $R^{O(poly(d))}$ can be computed in $O(nz(A)cdotlogn)$ time, and we also show that a $(1+epsilon)$-approximate solution to the $el{l}_p$ regression problem $mi{n}_{xin{R}^d}|Ax-b{|}_p$ can be computed in $O(nz(A)cdotlogn+poly(d)log(1/epsilon)/epsilo{n}^2)$ time. Moreover, we can improve the embedding dimension or equivalently the sample size to $O(d^{3+p/2}log(1/epsilon)/epsilo{n}^2)$ without increasing the complexity.

Recommendations

• A framework for robust subspace learning
• Subspace learning by \(\ell^0\)-induced sparsity
• Linear embedding by joint robust discriminant analysis and inter-class sparsity
• Sublinear Cost Low Rank Approximation via Subspace Sampling
• Fast, robust and non-convex subspace recovery
• Robust non-parametric regression via incoherent subspace projections
• Robust Subspace Discovery via Relaxed Rank Minimization
• Input sparsity time low-rank approximation via ridge leverage score sampling

Cited in

(42)

• Approximate Newton methods
• Quantile regression for large-scale data via sparse exponential transform method
• Sublinear update time randomized algorithms for dynamic graph regression
• Sparser Johnson-Lindenstrauss transforms
• An efficient algorithm for computing the approximate t-URV and its applications
• Estimating Leverage Scores via Rank Revealing Methods and Randomization
• Fast and accurate randomized algorithms for linear systems and eigenvalue problems
• Randomized Quaternion Singular Value Decomposition for Low-Rank Matrix Approximation
• ISLET: fast and optimal low-rank tensor regression via importance sketching
• Single pass spectral sparsification in dynamic streams
• Structural Convergence Results for Approximation of Dominant Subspaces from Block Krylov Spaces
• Erasure coding for fault-oblivious linear system solvers
• Randomized numerical linear algebra: Foundations and algorithms
• Sketched ridge regression: optimization perspective, statistical perspective, and model averaging
• Weighted SGD for \(\ell_p\) regression with randomized preconditioning
• Nearly tight oblivious subspace embeddings by trace inequalities
• Sharper bounds for regularized data fitting
• Scalable kernel \(k\)-means clustering with Nyström approximation: relative-error bounds
• Optimal subsampling for large‐sample quantile regression with massive data
• Spectral estimation from simulations via sketching
• \texttt{pylspack}: parallel algorithms and data structures for sketching, column subset selection, regression, and leverage scores
• Simpler is better: a comparative study of randomized pivoting algorithms for CUR and interpolative decompositions
• Tight Bounds for ℓ ₁ Oblivious Subspace Embeddings
• On approximating matrix norms in data streams
• Online row sampling
• Lower Memory Oblivious (Tensor) Subspace Embeddings with Fewer Random Bits: Modewise Methods for Least Squares
• Practical sketching algorithms for low-rank matrix approximation
• Tight bounds for \(\ell_p\) oblivious subspace embeddings
• On randomized sketching algorithms and the Tracy-Widom law
• Sampling Algorithms and Coresets for $\ell_p$ Regression
• Structured random sketching for PDE inverse problems
• Efficient bounds and estimates for canonical angles in randomized subspace approximations
• Sampling Lasso quantile regression for large-scale data
• New subset selection algorithms for low rank approximation: offline and online
• A User-Friendly Computational Framework for Robust Structured Regression with the L₂ Criterion
• scientific article; zbMATH DE number 7625175 (Why is no real title available?)
• Principled interpolation of Green's functions learned from data
• Streaming low-rank matrix approximation with an application to scientific simulation
• Tensor-structured sketching for constrained least squares
• Approximate sparse linear regression
• A count sketch maximal weighted residual Kaczmarz method for solving highly overdetermined linear systems
• Randomized algorithms in numerical linear algebra

Describes a project that uses

Uses Software

• LSRN

This page was built for publication: Low-distortion subspace embeddings in input-sparsity time and applications to robust linear regression

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5495779)