Présentation sur YouTube

Data-driven decision support systems are an important area of research with applications in diverse domains ranging from reconnaissance and surveillance for defence, risk assessment in business decisions to recommendation systems used every day.

We shall briefly discuss the various human factors involved in decision models for human-automation teams before proceeding to present our work. The focus of this talk is on an asymmetry inherent to a human-automation team, which is the workload dependence of human performance and the relative independence of automation performance.

A motivating scenario is a radar system tracking aircraft within its range, and classifying them as hostile or non-hostile based on parameters such as speed, altitude, direction, weapon signatures, etc. An important question is which targets must be referred to the human for review and final decision.

We consider a model for optimal decision referrals in human-automation teams performing binary classification tasks. Our key modeling assumption is that the human performance degrades with workload (i.e., the number of tasks referred to the human). We model the problem as a stochastic optimization problem. We show that when the workload of the human is pre-specified, it is optimal to myopically refer tasks which lead to the largest reduction in the conditional expected cost until the desired workload target is met. Based on this result we provide an algorithm to efficiently choose the subset of tasks to refer, for a general setting where there is no constraint on workload.