• Όλες οι κατηγορίες
• Ενδοπολυτεχνικές Ανακοινώσεις και Συζητήσεις
• Ακαδημαϊκές Ανακοινώσεις
• Ανακοίνωση παρουσίασης διπλωματικής εργασίας Αγγελιδάκη Α. - ΗΜΜΥ

Ανακοίνωση παρουσίασης διπλωματικής εργασίας Αγγελιδάκη Α. - ΗΜΜΥ

• 1

• Συντάχθηκε 25-02-2013 15:13 από Galateia Malandraki

  Email συντάκτη: gmalandraki<στο>tuc.gr

  Ενημερώθηκε: -

  Ιδιότητα: υπάλληλος ΑΡΜΗΧ.
  ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
  Τμήμα Ηλεκτρονικών Μηχανικών & Μηχανικών Υπολογιστών
  
  ΠΑΡΟΥΣΙΑΣΗ ΔΙΠΛΩΜΑΤΙΚΗΣ ΕΡΓΑΣΙΑΣ
  
  ΑΓΓΕΛΟΥ ΑΓΓΕΛΙΔΑΚΗ
  
  με θέμα
  
  “ Μάθηση Στρατηγικών για Εντοπισμό και Αποκατάσταση Βλαβών Δικτύου”
  “ Learning Strategies for Network Fault Detection and Remediation”
  
  Τετάρτη 27 Φεβρουαρίου 2013, 1μμ
  Αίθουσα Εργαστηρίου Intelligence (141-A14-2), Κτίριο Επιστημών, Πολυτεχνειούπολη
  
  Εξεταστική Επιτροπή
  
  Επίκουρος Καθηγητής Μιχαήλ Γ. Λαγουδάκης (επιβλέπων)
  Επίκουρος Καθηγητής Γεώργιος Χαλκιαδάκης
  Επίκουρος Καθηγητής Πολυχρόνης Κουτσάκης
  
  
  Abstract
  
  Network repair is a domain of growing significance. Viewing the network repair problem as a sequential decision problem, offers the opportunity to design autonomous agents able to implement minimum-cost fault detection and remediation policies through the execution of appropriate test and repair actions on the network. A repair policy consists of a sequence of test actions that help diagnose the faulty node of the network, followed by repair actions to restore it to proper operation with minimum cost. This thesis extends the Cost-Sensitive Fault Remediation (CSFR) model developed for such problems by introducing a minimax objective for planning and learning and applies the extended CSFR model and algorithms to the problem of network repair. The minimax objective, in contrast to the minimum expected cost objective of the original CSFR planning algorithm, aims to deliver safer policies which keep the maximum (worst) cost low. The minimax objective is complemented with three heuristics for resolving possible repair action ordering issues. The original planning algorithm and the three variations of the proposed planning algorithm are applied on fault detection and remediation problems on simulated networks of various sizes. To account for the computational limitations of the planning agent, an agent using instance-based reinforcement learning is developed. While the learning agent cannot guarantee optimality, it can handle much larger problems by exploiting data from interaction with the environment. Our results with both planning and learning indicate that the CSFR-based approach is able to handle network repair problems effectively. In several cases, the proposed minimax objective yields a maximum actual repair cost which is lower compared to the actual cost of the minimum expected cost objective.
  

  ---

  Συνημμένα:

  • 2013-02-27-Aggelidakis-Announcement.doc
    Μεταφορτώσεις: 125, Μέγεθος: 33 KB application/msword

• 1