ΣΕΜΙΝΑΡΙΟ ΤΜΗΜΑΤΟΣ & CEID SOCIAL HOUR: “Distributed tensor reconstruction for big data and applications” Θωμάς Παπαπαστεργίου, Υποψήφιος Διδάκτορας.

Εικόνα pintela
Κατηγορία: 

Σεμινάριο Τμήματος & CEID social hour

 

Ημερομηνία-χώρος: Παρασκευή 11 Μαΐου, 3μμ. Κτίριο Β (Αίθουσα Συνεδριάσεων - ισόγειο).

Ομιλητής: Θωμάς Παπαπαστεργίου, Υποψήφιος Διδάκτορας

Τίτλος: “Distributed tensor reconstruction for big data and applications”

Περίληψη: Data nowadays tend to be massive in volume and multi-dimensional in structure. Tensors (a.k.a. multi-dimensional arrays) are a natural way to represent efficiently such data. In many real-life applications data contain missing values due to temporary malfunction of sensors, costly experiments (e.g. seismic data collection) or detected outliers that have been removed prior to data analysis. Tensor decompositions (e.g.). Tucker or CANDECOMP/PARAFAC decomposition), as an extension of matrix decompositions and PCA in high order arrays, provide a valuable tool for multidimensional data analysis and mining. In this talk, a review of PARAFAC decomposition will be presented focusing on the challenges of scalable decompositions with missing values of very large tensors (orders up to 108x108x108). A distributed proximal method for dealing with this problem will be presented, which is based on solving local optimization problems rather than confronting the whole problem at once. At the end applications of the proximal algorithm in image completion and natural image annotation and classification will be presented.

Σύντομο βιογραφικό ομιλητή:

Thomas Papastergiou obtained his Bachelor from the Mathematics Department of University of Patras, his Master degree on “Mathematics of Computers and Decision Making” from the Departments of Mathematics and Computer Engineering and Informatics (CEID) of University of Patras, while currently he is a PhD student in the MDAKM laboratory at CEID with advisor Prof. Megalooikonomou. His research interests comprise multidimensional and distributed data analysis, big data, knowledgediscovery, tensors and machine learning.

Startup Growth Lite is a free theme, contributed to the Drupal Community by More than Themes.