Ph.D
Group : Learning and Optimization
Portfolio methods in uncertain contexts
Starts on 14/03/2013
Advisor : TEYTAUD, Olivier
Funding :
Affiliation : Université Paris-Saclay
Laboratory : LRI-TAO
Defended on 11/12/2015, committee :
Directeurs de thèse :
M. Olivier Teytaud, INRIA Saclay
M. Marc Schoenauer, INRIA Saclay
Rapporteurs :
M. Bruno Bouzy, Université Paris Descartes
Examinateurs :
M. Philippe Dague, Université Paris-Saclay
M. Simon Lucas, University of Essex
M. Petr Posik, Gerstner Laboratory
M. Günter Rudolph, University of Dortmund
Research activities :
Abstract :
The energy investments are difficult because of uncertainties. Some uncertainties can be modeled by the probabilities. But there are difficult issues such as the evolution of technology and the penalization of CO2, which can not be presented by probabilities. Also, in the traditional optimization of energy systems, disappointingly, the noise is often badly treated by deterministic management. This thesis focuses on applying noisy optimization to energy systems. This thesis concentrates in studying methods to handle noise, including using of resampling methods to improve the convergence rates; applying portfolio methods to noisy optimization in the continuous domain; applying portfolio methods to the energy investments and games, including the use of adversarial bandit algorithms to calculate the Nash equilibrium of two-player zero-sum matrix game and the use of "sparsity" to accelerate the computation of Nash equilibrium.
