MSc thesis of Kyriakos Sergis
Computational Aspects of the Braess Paradox
Supervisor: Dimitris Fotakis
In this thesis, we investigate the Braess paradox from a computational viewpoint. The motivation is to provide simple ways of improving network performance by exploiting the essence of the Braess's Paradox, namely the fact the network performance at equilibrium can be improved by edge removal. We first present approximation algorithms for the best subnetwork problem in random networks with linear latencies and polynomially many paths, each of polylogarithmic length. Moreover, we improve on the best known running time for the best subnetwork problem in certain classes of networks.
Defended: March 2, 2015.