Semi-Supervised Learning over Complex Networks

Loading...
Thumbnail Image
Journal Title
Journal ISSN
Volume Title
Perustieteiden korkeakoulu | Master's thesis
Date
2019-03-11
Department
Major/Subject
Complex Networks
Mcode
SCI3060
Degree programme
Master’s Programme in Life Science Technologies
Language
en
Pages
49 + 3
Series
Abstract
This work considers semi-supervised learning over network-structured datasets with an emphasis on modern convex optimization methods. As a case study, we investigate two specific variants of the Network Lasso problem: The Network Lasso and Logistic Network Lasso. We solve these using the Alternating Direction Method of Multipliers. Especially for the Logistic Network Lasso, we also give algorithms based on an inexact variant of ADMM and the primal-dual method. Our theoretical investigation is complemented with experiments conducted on artificial and real datasets.

Tämä työ käsittelee semi-valvottua oppimista verkko-rakenteissa, korostaen moderneja konveksi optimoinnin menetelmiä. Tapaustutkimuksena tarkastellaan kahta Network Lasso ongelman versiota: Network Lasso ja Logistic Network Lasso. Ratkaisemme nämä käyttämällä Alternating Direction Method of Multipliers menetelmää. Logistic Network Lasso ongelmalle johdamme myös epätäsmälliseen ADMM:ään ja primal-dual menetelmään perustuvia algoritmeja. Teoreettista tutkimustamme täydentävät keinotekoiseen sekä reaaliseen dataan perustuvat kokeilut.
Description
Supervisor
Jung, Alexander
Thesis advisor
Tran, Nguyen
Keywords
convex optimization, semi-supervised learning, complex networks, graph clustering
Other note
Citation