- 2 views
Monday, April 8, 2019 - 10:15 am
Storey Innovation Center (Room 2277)
Monday, April 8, 2019, in the Storey Innovation Center (Room 2277) from 10:15 am - 11:15 am.
Abstract: There are two fundamental strategies for finding effective policies to solve stochastic optimization problems and these are policy search and look-ahead policies. Look-ahead policies are mainly used in the context of sequential optimization problems in which the current decision impacts the future ones. There are several types of look-ahead policies depending on the type of the forecast, the problem structure and its dimensionality. In this talk, I will discuss two forms of look-ahead policies, direct look-aheads and value function approximations. In the context of direct look-aheads, we propose a new technique called Primal-Dual Monte Carlo Trees Search that utilizes sampled information relaxation upper bounds on potential actions, creating the possibility of “ignoring" parts of the tree that stem from highly suboptimal choices. This allows us to prove that despite converging to a partial decision tree in the limit, the recommended action from Primal-Dual MCTS is optimal. Then, I will discuss an approximate dynamic programming approach in the context of ride-sharing systems. We extract and prove important properties about the problem structure such as monotonicity and spatial correlation that provide efficient value function approximations.
Biography: Lina Al-Kanj is an Associate Research Scholar at the Operations Research and Financial Engineering Department at Princeton University. She received her PhD in Electrical and Computer Engineering from the American University of Beirut. Her research interests include optimal resource allocation and scheduling, stochastic optimization, dynamic programming and optimal learning with applications to energy, communication and transportation systems.