Συντάχθηκε 25-02-2021 15:43
Τόπος: Η παρουσίαση θα γίνει με τηλεδιάσκεψη
Σύνδεσμος τηλεδιάσκεψης
Έναρξη: 01/03/2021 10:30
Λήξη: 01/03/2021 11:30
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών
Πρόγραμμα Μεταπτυχιακών Σπουδών
ΠΑΡΟΥΣΙΑΣΗ ΜΕΤΑΠΤΥΧΙΑΚΗΣ ΕΡΓΑΣΙΑΣ
ΙΩΑΝΝΑ ΣΙΑΜΙΝΟΥ
θέμα
Stochastic Optimization on Tensor Factorization and Completion
Εξεταστική Επιτροπή
Καθηγητής Αθανάσιος Λιάβας (επιβλέπων)
Καθηγητής Γεώργιος Καρυστινός
Αναπληρωτής Καθηγητής Βασίλειος Σαμολαδάς
Abstract
We consider the problem of structured canonical polyadic decomposition (CPD). If the size of the problem is very big, then stochastic optimization approaches are viable alternatives to classical methods, such as Alternating Optimization (AO) and All-At-Once (AAO) optimization. We extend a recent stochastic gradient approach by employing an acceleration step (Nesterov momentum) in each iteration. We compare our approach with state-of-the-art alternatives, using both synthetic and real-world data, and find it to be very competitive. Furthermore, we examine the drawbacks of a parallel implementation of our accelerated stochastic algorithm and describe an alternative method that deals with these limitations. Finally, we propose an accelerated stochastic algorithm for the Nonnegative Tensor Completion problem and its parallel implementation via the shared–memory API OpenMP. Through numerical experiments, we test its efficiency in very large problems.
Meeting ID: 838 3360 3200
Password: 479572
