Αλγοριθμικές τεχνικές μάθησης πιθανοτικών κατανομών και εφαρμογές τους σε προβλήματα κοινωνικής επιλογής

Βλατάκης Γκαραγκούνης, Εμμανουήλ Βασίλειος; Vlatakis Gkaragkounis, Emmanouil Vasileios

dc.contributor.author	Βλατάκης Γκαραγκούνης, Εμμανουήλ Βασίλειος	el
dc.contributor.author	Vlatakis Gkaragkounis, Emmanouil Vasileios	en
dc.date.accessioned	2016-10-10T10:06:53Z
dc.date.available	2016-10-10T10:06:53Z
dc.date.issued	2016-10-10
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/43762
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.12572
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc/3.0/gr/	*
dc.subject	Θεωρία μάθησης	el
dc.subject	Εφαρμοσμένες πιθανότητες	el
dc.subject	Αλγόριθμοι κοινωνικής επιλογής	el
dc.subject	Πληθοψηφορία	el
dc.subject	Κατασκευή αραιών καλυμμάτων κατανομών	el
dc.subject	Computational learning theory	en
dc.subject	Computational social choice	en
dc.subject	Crowdvoting	en
dc.subject	Sparse cover in probability distributions	en
dc.subject	Applied probabilities	en
dc.title	Αλγοριθμικές τεχνικές μάθησης πιθανοτικών κατανομών και εφαρμογές τους σε προβλήματα κοινωνικής επιλογής	el
heal.type	bachelorThesis
heal.secondaryTitle	Design of probability distribution learning algorithms with applications in ploblems of computational social choice	en
heal.classification	Computer Science	el
heal.language	el
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2016-07-19
heal.abstract	In this thesis, we study probability distribution learning problems from a computational algorithmic perspective. We work in a natural PAC-style model of learning an unknown discrete probability distribution. In this framework, the learner is provided with the value of n and with independent samples drawn from the unknown distribution X. Using these samples, the learner must with probability at least 1−δ output a hypothesis distribution Xˆ such that the total variation distance dTV (X, Xˆ ) is at most ε, where ε, δ > 0 are accuracy and confidence parameters that are provided to the learner. We present the previous work on that framework for classical classes of probability distributions i.e monotone, log-concave, unimodal distributions giving tight upper and lower sample bounds. We focus on a new algorithmic technique in distribution learning, the construction of covers, that are sparse in cardinality and dense in metric space of a class of distributions. The method exploits that structure in order to design efficient in time and sample complexity learning algorithms. We study the seminal work of [DP10] and [DDS12] on the Poisson Binomial Distribution, the sum of n independent Bernoulli. We develop then a similar approach on an gradually emerging field of computational social choice, crowdvoting. Based on the famous Mallow noise Model in which every voter is an estimator of the social ground truth, we present our work for the Kemeny and Plurality rule extending the previous research results.	en
heal.advisorName	Φωτάκης, Δημήτριος	el
heal.committeeMemberName	Παγουρτζής, Αριστείδης	el
heal.committeeMemberName	Λουλάκης, Μιχαήλ	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών. Τομέας Τεχνολογίας Πληροφορικής και Υπολογιστών	el
heal.academicPublisherID	ntua
heal.numberOfPages	134 σ.	el
heal.fullTextAvailability	true