Learning Poisson Binomial Distributions with Differential Privacy

Postgraduate Thesis uoadl:1332104 349 Read counter

Unit:
Κατεύθυνση Λογική και Θεωρία Αλγορίθμων και Υπολογισμού
Library of the School of Science
Deposit date:
2017-03-01
Year:
2017
Author:
Giannakopoulos Agamemnon
Supervisors info:
Φωτάκης Δημήτριος, Επίκουρος Καθηγητής, τμήμα Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών του Ε.Μ.Π.
Ευστάθιος Ζάχος, Ομότιμος Καθηγητής, τμήμα Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών του Ε.Μ.Π.
Αριστείδης Παγουρτζής, Αναπληρωτής Καθηγητής, τμήμα Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών του Ε.Μ.Π.
Original Title:
Learning Poisson Binomial Distributions with Differential Privacy
Languages:
English
Translated title:
Learning Poisson Binomial Distributions with Differential Privacy
Summary:
This thesis tries to leverage two major research areas. The first area concerns the Distribution Learning area and the second the Differential Privacy. More specific, given a highly efficient algorithm which learns with ε-accuracy a Poisson Binomial distribution we try to study its Differential Privacy property. We show that if the algorithm is close to a (n,k)-Binomial form the algorithm is differential private. If the PBD is close to a k-Sparse form the algorithm's privacy depends on PBD cardinality
Main subject category:
Science
Other subject categories:
Mathematics
Keywords:
Learning, Poisson, Binomial, Distribution, Differential, Privacy
Index:
No
Number of index pages:
0
Contains images:
Yes
Number of references:
51
Number of pages:
67
Thesis v01.pdf (2 MB) Open in new window