Experimentally disambiguating models of sensory cue integration

Abstract

Sensory cue integration is one of the primary areas in which a normative mathematical framework has beenused to define the “optimal” way in which to makedecisions based upon ambiguous sensory informationand compare these predictions to behavior. Theconclusion from such studies is that sensory cues areintegrated in a statistically optimal fashion. However,numerous alternative computational frameworks existby which sensory cues could be integrated, many ofwhich could be described as “optimal” based ondifferent criteria. Existing studies rarely assess theevidence relative to different candidate models,resulting in an inability to conclude that sensory cuesare integrated according to the experimenter’s preferredframework. The aims of the present paper are tosummarize and highlight the implicit assumptions rarelyacknowledged in testing models of sensory cueintegration, as well as to introduce an unbiased andprincipled method by which to determine, for a givenexperimental design, the probability with which apopulation of observers behaving in accordance withone model of sensory integration can be distinguishedfrom the predictions of a set of alternative models