Velibor Misic

 

Faculty Candidate Velibor Misic [Massachusetts Institute of Technology] 
Title Data, Models and Decisions in Large-Scale Stochastic Optimization Problems
Date & Time Friday, February 12, 2016 at 10:30am
Place UCLA Anderson School of Management 
Room D310

 

Abstract
Modern business decisions exceed human decision making ability: they are large-scale, their outcomes are uncertain, and they are made in multiple stages. At the same time, firms have increasing access to data and models that allows them to understand the effects of different decisions. Faced with such complex decisions and increasing access to data and models, a fundamental question that arises is: how do we transform data and models into effective decisions? In this talk, we address this question in the context of two important problems: the problem of making tactical assortment decisions - a key problem in operations management - and the problem of dynamically controlling large-scale stochastic systems - a key problem in operations research.

In the first part of the talk, we propose a new two-step method for transforming limited customer transaction data into effective assortment decisions. The approach involves estimating a ranking-based model of choice from data by solving a large-scale linear optimization problem, and then solving a practically tractable mixed-integer optimization problem to obtain an assortment decision from the estimated predictive model. Using synthetic transaction data, we show that the approach is scalable, leads to accurate predictions and effective decisions that outperform alternative parametric and non-parametric approaches.
In the second part of the talk, we propose a new solution method for a general class of Markov decision processes (MDPs) called decomposable MDPs. Decomposable MDPs are problems where the problem data decomposes along the components of the system. We propose a novel linear optimization formulation that directly exploits the decomposable nature of the problem data to obtain a heuristic for the true problem. We show that the formulation is theoretically stronger than alternative proposals and provide extensive numerical evidence for the strength of the method in multiarmed bandit and optimal stopping problems. 

Velibor Misic