Unit:
Κατεύθυνση Ηλεκτρονική και Ραδιοηλεκτρολογία (Ρ/Η, με πρόσθετη εξειδίκευση στις Τηλεπικοινωνίες και στην επεξεργασία και διοίκηση της Πληροφορίας)Library of the School of Science
Author:
Λαμπίρης Ελευθέριος
Supervisors info:
Άρης Μουστάκας
Original Title:
Reaching an Equilibrium: Learning and Game Theory
Translated title:
Φτάνοντας σε Ισορροπία: Μάθηση και Θεωρία Παιγνίων
Summary:
This thesis is concerned with studying various learning algorithms applied in a
game theoretic environment and most importantly in how these can help optimise
non-cooperative wireless networks.
The two algorithmic families are multi-armed bandits and evolutionary game
theory and the goal is the study of their time evolution.
In this thesis we will analyse the common background between the two
algorithmic families and we will focus on the required parameters under which
these algorithms converge to a Nash, pure or mixed, equilibrium. To evaluate
the algorithms we will use utility functions designed to reduce the consumed
power in nodes - transmitters in a wireless network. Finally, the role of the
quantity of feedback in learning will be studied.
Keywords:
Evolutionary game theory, Distributed resource management, Limited feedback, Multi-armed bandits, Wireless networks
