Discrepancy Minimization via Regularization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Pesenti, Lucas, Vladu, Adrian
Format: Preprint
Published: 2022
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918446176927744
author Pesenti, Lucas
Vladu, Adrian
author_facet Pesenti, Lucas
Vladu, Adrian
contents We introduce a new algorithmic framework for discrepancy minimization based on regularization. We demonstrate how varying the regularizer allows us to re-interpret several breakthrough works in algorithmic discrepancy, ranging from Spencer's theorem [Spencer 1985, Bansal 2010] to Banaszczyk's bounds [Banaszczyk 1998, Bansal-Dadush-Garg 2016]. Using our techniques, we also show that the Beck-Fiala and Komlos conjectures are true in a new regime of pseudorandom instances.
format Preprint
id arxiv_https___arxiv_org_abs_2211_05509
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Discrepancy Minimization via Regularization
Pesenti, Lucas
Vladu, Adrian
Data Structures and Algorithms
Discrete Mathematics
We introduce a new algorithmic framework for discrepancy minimization based on regularization. We demonstrate how varying the regularizer allows us to re-interpret several breakthrough works in algorithmic discrepancy, ranging from Spencer's theorem [Spencer 1985, Bansal 2010] to Banaszczyk's bounds [Banaszczyk 1998, Bansal-Dadush-Garg 2016]. Using our techniques, we also show that the Beck-Fiala and Komlos conjectures are true in a new regime of pseudorandom instances.
title Discrepancy Minimization via Regularization
topic Data Structures and Algorithms
Discrete Mathematics
url https://arxiv.org/abs/2211.05509