Event details from: Business Analytics Seminar: Orit Davidovich (IBM)

Date
12 December 2022
Time
12:00 – 13:00

Introduction

Mathematical optimization models can improve decision making in a wide variety of industrial settings, such as process control in manufacturing. Creating optimization models, however, requires domain as well as optimization expertise, thereby limiting the use of this methodology. One way to overcome this gap is by utilizing historical data to learn (parts of) the optimization model (i.e., objectives or constraints). However, as historical decisions may be suboptimal (though informed), the data used for training may not be representative of highly optimal (or severely suboptimal) operation regions, resulting in critical inaccuracies.

In this talk, I will introduce a formal approach for ensuring the quality of optimization solutions coming from learned optimization models. This approach combines a new metric for solution quality, utilizing Gaussian Processes, together with iterative model augmentation to improve solution quality. I will demonstrate the accuracy and usefulness of this approach over current state of the art on the World Food Program food basket optimization problem. This forms part of joint work with my colleagues at IBM Research, Segev Wasserkrug, Parikshit Ram, and Dharmashankar Subramanian.

General information

This seminar will be organised in a hybrid setup. If you are interested in joining this seminar, please send an email to the secretariat of Amsterdam Business School at secbs-abs@uva.nl.