Συντάχθηκε 14-09-2012 10:17
από Galateia Malandraki
Email συντάκτη: gmalandraki<στο>tuc.gr
Ενημερώθηκε:
-
Ιδιότητα: υπάλληλος ΑΡΜΗΧ.
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Τμήμα Ηλεκτρονικών Μηχανικών & Μηχανικών Υπολογιστών
ΠΑΡΟΥΣΙΑΣΗ ΔΙΠΛΩΜΑΤΙΚΗΣ ΕΡΓΑΣΙΑΣ
Αλεβίζου Παναγιώτη
με θέμα
“Γράφοι Παραγόντων: Θεωρία και Εφαρμογές”
“Factor Graphs: Theory and Applications”
Τετάρτη 19 Σεπτεμβρίου 2012, 5μ.μ.
Αίθουσα 137.Π39, Κτίριο Επιστημών, Πολυτεχνειούπολη
Εξεταστική Επιτροπή
Άγγελος Μπλέτσας, Επίκουρος Καθηγητής (επιβλέπων)
Αθανάσιος Λιάβας , Καθηγητής
Καρυστινός Γιώργιος, Επίκουρος Καθηγητής
Abstract
Factor graphs (FGs) represent graphically the factorization of a global function
into a product of local sub-functions. The global function is usually a multi-variable probability density function (pdf),
where the calculation of a marginal pdf is usually intractable. The sum-product algorithm (SPA)
is applied on the FG through message-passing, i.e. exchange of functions, between the FG nodes in a distributed way;
the output is a marginal pdf with respect to a variable of interest. Factor
graph theory has several applications in many interdisciplinary fields, such as error correction coding theory,
detection and estimation, wireless networking, artificial intelligence and many others.
This thesis
provides the basic theoretical background in a tutorial way, from first principles. Furthermore,
specific FG applications found in the literature are presented.
Specifically, coding problems (LDPC, convolutional and parallel concatenated Turbo codes), Bayesian estimation
(in the context of network localization) and wireless multi-hop networking (in the context of time scheduling)
are analyzed within the FG framework. In all cases, the respective graph, the associated SPA,
the message-passing scheduling and the final output are thoroughly presented. The power of FGs as a distributed inference
tool is vividly demonstrated.
Συνημμένα:
-
Diplomatiki-Anakoinwsi_AlevizosPanos.doc
Μεταφορτώσεις: 168,
Μέγεθος: 42 KB application/msword
