Recovering Simultaneously Structured Data via Non-Convex Iteratively Reweighted Least Squares

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Kümmerle, Christian, Maly, Johannes
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914644567785472
author Kümmerle, Christian
Maly, Johannes
author_facet Kümmerle, Christian
Maly, Johannes
contents We propose a new algorithm for the problem of recovering data that adheres to multiple, heterogeneous low-dimensional structures from linear observations. Focusing on data matrices that are simultaneously row-sparse and low-rank, we propose and analyze an iteratively reweighted least squares (IRLS) algorithm that is able to leverage both structures. In particular, it optimizes a combination of non-convex surrogates for row-sparsity and rank, a balancing of which is built into the algorithm. We prove locally quadratic convergence of the iterates to a simultaneously structured data matrix in a regime of minimal sample complexity (up to constants and a logarithmic factor), which is known to be impossible for a combination of convex surrogates. In experiments, we show that the IRLS method exhibits favorable empirical convergence, identifying simultaneously row-sparse and low-rank matrices from fewer measurements than state-of-the-art methods. Code is available at https://github.com/ckuemmerle/simirls.
format Preprint
id arxiv_https___arxiv_org_abs_2306_04961
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Recovering Simultaneously Structured Data via Non-Convex Iteratively Reweighted Least Squares
Kümmerle, Christian
Maly, Johannes
Machine Learning
Information Theory
Optimization and Control
We propose a new algorithm for the problem of recovering data that adheres to multiple, heterogeneous low-dimensional structures from linear observations. Focusing on data matrices that are simultaneously row-sparse and low-rank, we propose and analyze an iteratively reweighted least squares (IRLS) algorithm that is able to leverage both structures. In particular, it optimizes a combination of non-convex surrogates for row-sparsity and rank, a balancing of which is built into the algorithm. We prove locally quadratic convergence of the iterates to a simultaneously structured data matrix in a regime of minimal sample complexity (up to constants and a logarithmic factor), which is known to be impossible for a combination of convex surrogates. In experiments, we show that the IRLS method exhibits favorable empirical convergence, identifying simultaneously row-sparse and low-rank matrices from fewer measurements than state-of-the-art methods. Code is available at https://github.com/ckuemmerle/simirls.
title Recovering Simultaneously Structured Data via Non-Convex Iteratively Reweighted Least Squares
topic Machine Learning
Information Theory
Optimization and Control
url https://arxiv.org/abs/2306.04961