Context Dependency of Pattern-Category Learning

Abstract

Despite its widely acknowledged importance context has remained a relatively vague concept in vision research. Previous approaches regard context primarily as a determinant for the interpretation of sensory information on the basis of previously acquired knowledge. In this paper we propose a complementary perspective, by showing that context also specifically affects learning, that is the acquisition of knowledge and the way in which such knowledge is mentally represented. In two pattern-category learning experiments we explored how complementary manipulations of context affect learning performance and generalization. In both experiments, generalization performance was measured as the ability to transfer acquired class knowledge to the contrast-inverted versions of the learning patterns. We then modelled the behavioural data in terms of evidence-based classification. Such an analysis allows to reconstruct combinations of non-relational and relational pattern attributes that provide potential solutions of a given classification problem. We show that ‘context’ in category learning affects the search within the search space of attribute combinations which underlie the production rules for the categories. Our results suggest a novel, context-based explanation for well-known phenomena of contrast-invariance in visual perception.