Information Elicitation for Improving Optimization: Applications in Scheduling Problems

When we work on a practical scheduling task, we usually do not have complete knowledge of the related resources and constraints. For example, when scheduling a conference, we may not know the exact sizes of available rooms or equipment needs of some speakers. The task of constructing a schedule based on incomplete data gives rise to several related problems, including the representation of uncertainty, efficient search for schedules, and elicitation of additional data that help to reduce uncertainty. In this work we introduce a new information elicitation approach aiming to resolve uncertainty in order to increase quality of optimization while keeping the number of questions the user has to answer to a minimum. The approach differs from other approaches in terms of working with a continuous domain with a large number of uncertain variables, not having a need for bootstrapping, tight integration with the optimization process and integration of multiple approaches to elicitation.