Democratizing Optimization Modeling: Status, Challenges, and Future Directions

Segev Wasserkug
IBM Research - Israel

Using optimization to make better business decisions can provide significant value to many enterprises in a variety of industries. A good optimization model needs to both capture the business problem in precise terms, and to be specified in a way that it can be efficiently solved. Therefore, traditionally, the creation of such models requires a lengthy modeling process by people with rare optimization modeling expertise. The rarity of the skills and the length of time required are significant inhibitors to the widespread use of optimization.
In this talk, I will discuss how new AI techniques seem to be on the brink of revolutionizing optimization model creation, by allowing almost anyone to create such models. I will discuss techniques such as constraint learning, learning to optimize, and large language models such as GPT, and how they use a combination of data and knowledge driven techniques to radically simplify optimization model creation. I will also present new issues introduced by the use of these techniques, and research challenges in overcoming these issues and fulfilling the vision of democratizing optimization modeling. I will also discuss in detail two works I am involved in to address some of these gaps: how to ensure the quality of solutions obtained by optimization models with constraint learning, and a spreadsheet driven optimization modeling approach intended for business analysts.

Presentation (PDF File)

Back to Artificial Intelligence and Discrete Optimization