Thu.3 13:30–14:45 | H 3006 | ROB
.

Dynamic Optimization Under Data Uncertainty

Chair: Omid Nohadani Organizer: Omid Nohadani
13:30

Sebastian Pokutta

joint work with Andreas Bärmann, Alexander Martin, O. Schneider

From Robust Optimization to Online Inverse Optimization

Robust optimization can be solved via online learning requiring only access to the nominal problem for a given uncertainty realization. We show that a similar perspective can be assumed for an online learning approach to online inverse optimization, even for discrete and non-convex decisions. The decision maker observes an expert's decisions over time and learns an objective function that is equivalent to the expert's private objective. We present a new framework for inverse optimization through online learning that goes beyond duality approaches and demonstrate its real-world applicability.

13:55

Eojin Han

joint work with Omid Nohadani, Chaithanya Bandi

Robust Periodic-Affine Policies for Multiperiod Dynamic Problems

We introduce a new class of adaptive policies called periodic-affine policies, that allows a decision maker to optimally manage and control large-scale newsvendor networks in the presence of uncertain demand without distributional assumptions. These policies model many features of the demand such as correlation, and can be generalized to multi-product settings and multi-period problems. This is accomplished by modeling the uncertain demand via sets. We provide efficient algorithms and demonstrate their advantages on the sales data from one of India’s largest pharmacy retailers.

14:20

Dan Iancu

joint work with Nikos Trichakis, Do-Young Yoon

Monitoring With Limited Information

We consider a system with an evolving state that can be stopped at any time by a decision maker (DM), yielding a state-dependent reward. The DM observes the state only at limited number of monitoring times, which he must choose, in conjunction with a stopping policy. We propose a robust optimization approach, whereby adaptive uncertainty sets capture the information acquired through monitoring. We consider two versions of the problem, static and dynamic and show that, under certain conditions, the same reward is achievable under either static or dynamic monitoring.