Efficient geometric random walks for high-dimensional sampling from convex bodies

Abstract

High-dimensional sampling is a fundamental problem with many applications in science and engineering. Several problems are computationally hard for general dimension, and therefore, a great effort has been devoted to randomized approximation algorithms based on sampling that addresses those problems in polynomial time. In this thesis, we present algorithmic, complexity, and implementation results on the problem of sampling points from a log-concave distribution restricted to a convex polytope –the feasible region of a linear program– or a spectrahedron –the feasible region of a semidefinite program (SDP). We develop practical Multiphase Monte Carlo algorithms to address the problem of approximating the volume of a convex body, sampling from the flux space of metabolic networks in Systems Biology, and estimating the optimal solution of an SDP. The corresponding implementations scale up to thousands or hundred dimensions, depending on the problem. We use those methods and develop new geo ...
show more

All items in National Archive of Phd theses are protected by copyright.

DOI
10.12681/eadd/51013
Handle URL
http://hdl.handle.net/10442/hedi/51013
ND
51013
Alternative title
Αποδοτικοί γεωμετρικοί τυχαίοι περίπατοι για τη δειγματοληψία από υψηλών διαστάσεων κυρτά σώματα
Marches aléatoires géométriques efficaces pour un échantillonnage de grande dimension à partir d'ensembles convexes
Author
Chalkis, Apostolos (Father's name: George)
Date
2021
Degree Grantor
National and Kapodistrian University of Athens
Committee members
Εμίρης Ιωάννης
Ζησιμόπουλος ασίλειος
Γιαννόπουλος Απόστολος
Joswig Michael
Φωτάκης Δημήτριος
Γουνόπουλος Δημήτριος
Δαλαμάγκας Θεόδωρος
Discipline
Natural SciencesComputer and Information Sciences ➨ Computer science, theory and methods
Keywords
Efficient geometric random walks for high-dimensional sampling from convex bodies; Apostolos Chalkis
Country
Greece
Language
English
Description
im., tbls.
Usage statistics
VIEWS
Concern the unique Ph.D. Thesis' views for the period 07/2018 - 07/2023.
Source: Google Analytics.
ONLINE READER
Concern the online reader's opening for the period 07/2018 - 07/2023.
Source: Google Analytics.
DOWNLOADS
Concern all downloads of this Ph.D. Thesis' digital file.
Source: National Archive of Ph.D. Theses.
USERS
Concern all registered users of National Archive of Ph.D. Theses who have interacted with this Ph.D. Thesis. Mostly, it concerns downloads.
Source: National Archive of Ph.D. Theses.