Modeling Eye Movement Sequences Using Conceptual Clustering Techniques

摘要

An algorithm for clustering noisy continuous numeric data was developed in a learning system called 2DCG (two dimensional cluster generalization). The 2DCG system operates in a two-dimensional space, but a general system could operate in an N-dimensional space. The objective of the system was to learn a set of rules which modeled human observers (in the application presented here, this model predicted changes in the eye position of human observers during a visual monitoring task). The rule set had to be complete, consistent, and nonredundant, while minimizing the number and maximizing the generality of the rules. The development of this model and its performance in accounting for noisy data are described. Keywords: Artificial intelligence, Concept learning, Conceptual clustering, Cognitive modeling, Eye movements, Grouping, Rule-based systems, Segmentation, Visual monitoring. (sdw)