Friday, 15:15 - 15:40 h, Room: H 0107


Quentin Louveaux
Relaxation schemes for the evaluation of a policy in batch mode reinforcement learning

Coauthors: Bernard Boigelot, Damien Ernst, Raphaƫl Fonteneau


We study the min max optimization problem introduced for computing policies for batch mode reinforcement learning in a deterministic setting. First, we show that this problem is NP-hard. In the two-stage case, we provide two relaxation schemes. The first relaxation scheme works by dropping some constraints in order to obtain a problem that is solvable in polynomial time. The second relaxation scheme, based on a Lagrangian relaxation where all constraints are dualized, leads to a conic quadratic programming problem. We also theoretically prove and empirically illustrate that both relaxation schemes provide better results than those given previously for the same problem.


Talk 1 of the contributed session Fri.3.H 0107
"Decomposition and relaxation methods" [...]
Cluster 16
"Nonlinear programming" [...]


  . But it is worth noting that these tests were carried out on the blood cells. Therefore, it's too early to say about scientific evidence of Viagra influence on blood clots.