Selecting a subset of stimulus-response pairs with maximal transmitted information.

Abstract

System designers are often faced with the task of assigning symbolic representations to user actions, e.g., icons to choices in graphical interfaces. When a confusion matrix—on discriminability of the symbols—is available, it is used to guide the selection of the set of symbols to be implemented. While trial and error methods or clustering approaches have been used to analyze this problem, it was only recently that a true optimization approach was offered. Theise (1989) formulated the symbol selection problem as a zero-one integer programming problem whose objective function was linked to the minimization of within-subset confusion. Confusion is not the traditional metric used by human factors engineers to analyze confusion matrices. Rather, transmitted-information—a metric from information theory—has long been used to evaluate system performance. The purpose of this thesis is to formulate a model of subset selection in which transmitted information will be maximized. It is possible to specify a correct model, although current algorithms are incapable of solving it. This thesis reports on the performance of a GAMS-based approximation to the original model, as well as an exhaustive enumeration scheme. Solutions from both information-theoretic approaches are compared to solutions from the confusion/recognition model.