Συντάχθηκε 10-10-2012 09:12
από Galateia Malandraki
Email συντάκτη: gmalandraki<στο>tuc.gr
Ενημερώθηκε:
-
Ιδιότητα: υπάλληλος ΑΡΜΗΧ.
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Τμήμα Ηλεκτρονικών Μηχανικών & Μηχανικών Υπολογιστών
ΠΑΡΟΥΣΙΑΣΗ ΔΙΠΛΩΜΑΤΙΚΗΣ ΕΡΓΑΣΙΑΣ
Ιωακείμ Πέρρος
με θέμα
“Soft co-clustering σε Map-Reduce μέσω κατανεμημένης Παλινδρόμησης Αραιού Πίνακα”
“Soft co-clustering in Map-Reduce using distributed Sparse Matrix Regression”
Τετάρτη 10 Οκτωβρίου 2012, 6μμ
Αίθουσα 2041, Κτίριο Επιστημών, Πολυτεχνειούπολη
Εξεταστική Επιτροπή
Καθηγητής Μίνως Γαροφαλάκης (επιβλέπων)
Επίκουρος καθηγητής Αντώνιος Δεληγιαννάκης
Επίκουρος καθηγητής Μιχαήλ Λαγουδάκης
Abstract
Modern technological and scientific advancements coupled with the massive increase in computing power and data storage capacity have given rise to a different paradigm of computer science. Often hailed as the Fourth Paradigm, it encompasses systems and techniques geared towards the analysis of and extraction of actionable knowledge from the abundance of generated data. Co-cluster analysis is an example of such useful knowledge extraction that describes the process of discovering concepts of a data matrix emerging from the correlations of row and column subsets.
This thesis studies the design and development of a co-clustering algorithm targeted for big data analysis. Our implementation is based on the Map-Reduce model and its open-source implementation, Hadoop. In contrast to earlier work, we approach Map-Reduce co-clustering through distributed optimization techniques exploiting the idea of Sparse Matrix Regression (SMR) which can define "soft" co-clusters capturing important correlations and masking noise in the input data. We give novel formulations of the different SMR subproblems in a parallelizable way over the Map-Reduce model of computation, and present an experimental study on real-life data that establishes the accuracy and scalability of our approach, as well as its ability to consistently discover new, robust concepts in large data collections.
Συνημμένα:
-
Diplomatiki-Anakoinwsi_Perros.doc
Μεταφορτώσεις: 179,
Μέγεθος: 44 KB application/msword
