סמינר בניהול טכנולוגיה ומערכות מידע

Prof. Maytal Saar-Tsechansky

McCombs School of Business

,    ,The University of  Texas at Austin

Abstract

Traditionally, information acquisition policies aim to intelligently select information used by predictive models.  This work has focused primarily on policies that aim to improve a single classification model through cost-effective information acquisitions. However, many important business decisions do not correspond to the literature’s focus on a single classification model. Rather, common repetitive decisions rely on inferences from multiple predictive models of different kinds, including regression and classification models. This challenge poses some interesting questions: How can we enable different kinds of predictive modeling tasks to jointly reason about information-acquisition opportunities to improve the decisions they inform? How should knowledge (or uncertainty) about the decisions the models inform what models it would be most beneficial to improve through information acquisition? I will present a new approach that aim to address these and related questions along with empirical results on common decision tasks, including direct marketing and sales tax non-compliance.