Mar 16, 2019

PhD Course on Choice-Based Optimization

Ph.D. Course on Choice-Based Optimization at Hamburg University 22nd to 25th of July 2019. The course is organized by the German Academic Association for Business Research within the ProDok PhD course program.

Choice-based optimization in short: integration of machine learning approaches for customer choices in mixed integer programs for smart decision making ... some would call it artificial intelligence ...

Scope and Outline:
Demand is an important quantity in many optimization problems, such as revenue management and supply chain management, for example. While some part of the literature considers demand as deterministic some reference assume demand to be stochastic. However, beyond stochasticity, demand usually depends on “supply” (price and availability of products, for example) which in turn is decided on in the optimization model. Hence, demand is endogenous to the optimization problem. Apart from revenue management and assortment optimization, most reference neglect demand endogeneity. Further, we often find aggregate demand models in optimization models. Choice-based optimization (CBO) is about to overcome these shortcomings. CBO merges discrete choice models with linear (mixed integer) programs. Discrete choice models (DCM) are matured in analyzing and predicting individual demand (i.e., disaggregate demand). These models are theoretically sound (based on utility maximization) and flexible. They are applied by both - practitioners and researchers - for more than four decades in various fields like transport, marketing, and consumer research, energy, and health care, for example. At its heart, DCM describe the choice probabilities of individuals selecting an alternative from a set of available alternatives (smartphones, for example). CBO determines (i) the availability of the alternatives (alternative selection problem) and/or (ii) the attributes of the alternatives (attribute problem), i.e., the decision variables determine the availability of alternatives and/or the shape of the attributes. As such CBO decisions determine demand derived from choice probabilities. Unfortunately, DCM come at the cost of high non-linearity and sometimes even non-closed form of the choice probabilities. In this seminar, we discuss various approaches to deal with these issues (non-linearity and non-closed form). We present CBO applications to location planning, supply chain management, product portfolio planning, and revenue management. We provide an outlook for future research - and collaboration - to further develop the field of choice-based optimization.