A learning-enhanced projection method for solving convex feasibility problems

Janosch Rieger

doi:10.3934/dcdsb.2021054

A learning-enhanced projection method for solving convex feasibility problems

Janosch Rieger

School of Mathematics

Research output: Contribution to journal › Article › Research › peer-review

Abstract

We propose a generalization of the method of cyclic projections, which uses the lengths of projection steps carried out in the past to learn about the geometry of the problem and decides on this basis which projections to carry out in the future. We prove the convergence of this algorithm and illustrate its behavior in a first numerical study.

Original language	English
Pages (from-to)	555-568
Number of pages	14
Journal	Discrete and Continuous Dynamical Systems - Series B
Volume	27
Issue number	1
DOIs	https://doi.org/10.3934/dcdsb.2021054
Publication status	Published - Jan 2022

Keywords

Acceleration by learning
Algebraic reconstruction technique
Convex feasibility problems
Kaczmarz method
Method of alternating projections

Access to Document

10.3934/dcdsb.2021054Licence: Unspecified

Cite this

@article{b5310ad85d0d4b55a088cc9b0f87d6b4,

title = "A learning-enhanced projection method for solving convex feasibility problems",

abstract = "We propose a generalization of the method of cyclic projections, which uses the lengths of projection steps carried out in the past to learn about the geometry of the problem and decides on this basis which projections to carry out in the future. We prove the convergence of this algorithm and illustrate its behavior in a first numerical study.",

keywords = "Acceleration by learning, Algebraic reconstruction technique, Convex feasibility problems, Kaczmarz method, Method of alternating projections",

author = "Janosch Rieger",

year = "2022",

month = jan,

doi = "10.3934/dcdsb.2021054",

language = "English",

volume = "27",

pages = "555--568",

journal = "Discrete and Continuous Dynamical Systems - Series B",

issn = "1531-3492",

publisher = "American Institute of Mathematical Sciences (AIMS)",

number = "1",

}

TY - JOUR

T1 - A learning-enhanced projection method for solving convex feasibility problems

AU - Rieger, Janosch

PY - 2022/1

Y1 - 2022/1

N2 - We propose a generalization of the method of cyclic projections, which uses the lengths of projection steps carried out in the past to learn about the geometry of the problem and decides on this basis which projections to carry out in the future. We prove the convergence of this algorithm and illustrate its behavior in a first numerical study.

AB - We propose a generalization of the method of cyclic projections, which uses the lengths of projection steps carried out in the past to learn about the geometry of the problem and decides on this basis which projections to carry out in the future. We prove the convergence of this algorithm and illustrate its behavior in a first numerical study.

KW - Acceleration by learning

KW - Algebraic reconstruction technique

KW - Convex feasibility problems

KW - Kaczmarz method

KW - Method of alternating projections

UR - http://www.scopus.com/inward/record.url?scp=85121919583&partnerID=8YFLogxK

U2 - 10.3934/dcdsb.2021054

DO - 10.3934/dcdsb.2021054

M3 - Article

AN - SCOPUS:85121919583

SN - 1531-3492

VL - 27

SP - 555

EP - 568

JO - Discrete and Continuous Dynamical Systems - Series B

JF - Discrete and Continuous Dynamical Systems - Series B

IS - 1

ER -

A learning-enhanced projection method for solving convex feasibility problems

Abstract

Keywords

Access to Document

Other files and links

Cite this