Toward an Eye-Tracking Adaptive Interface for Interactive Evolutionary Computation

Abstract

Interactive Evolutionary Computation (IEC) refers to optimization methods in which human evaluations are integrated into the search process [14]. This approach is particularly relevant when the quality of a solution cannot be defined through objective mathematical criteria alone, because it depends on subjective preferences, perceptual judgments, or implicit mental representations [9]. For instance, a user may have a mental representation of an ideal product, such as a piece of clothing or a graphical interface, without being able to explicitly formalize all the criteria defining this preference. However, IEC systems are limited by the number of evaluations that users can perform before cognitive fatigue, attentional overload, or disengagement occur [9]. Traditional evolutionary algorithms often require many evaluations to converge, whereas human users have limited cognitive resources. This creates a central challenge for human-algorithm interaction: supporting algorithmic convergence while preserving acceptable cognitive conditions for the user. This work-in-progress paper presents an experimental protocol comparing a fixed display with an eye-tracking-based adaptive display. The adaptive interface adjusts the number of displayed items according to the user’s estimated cognitive state. The study aims to assess whether this adaptation can reduce cognitive load while preserving convergence toward a target item, using measures such as recognition performance, search time, pupil dilation, workload, and number of displayed items.