Abstract
Visual category learning (VCL) is a cognitive process that involves identifying features shared by objects from the same category as well as features that discriminate objects from different categories even when these are not salient or when the feedback provided by supervision is ambiguous. This requires overcoming challenges both at the perceptual level and at the induction and decision-making level. To date, the interaction between these two systems has not been systematically tested. In the current study subjects learned to categorize novel 3D objects. In each VCL task objects varied along three feature dimensions, of which one was relevant for categorization and the other two were irrelevant. In one experimental condition, all visual features (both the relevant one and irrelevant ones) were highly salient and differences in each feature were easily perceived. In a second condition, all three visual features had low salience, and consequently changes along each feature dimension were hard to detect. We show that when VCL is done with highly informative supervision in which each trial is sufficient for learning the categorization rule, people are capable of learning this rule effectively within small number of learning trials for both low and high salience visual features. On the other hand, when each trial does not independently reveal the relevant feature, but it can be learned by accumulating evidence across trials, subjects can learn the categorization rule only when the features are highly visually salient. These data suggest that when the differences between objects of different categories are not visually salient, people fail to accumulate information across multiple learning trials, and therefore do not learn the categorization rule under ambiguous supervision. Overall, our data suggest that VCL can be done effectively either under conditions of ambiguous feedback information or poor feature saliency, but not when simultaneously facing both challenges.
Research is supported by RO1EY019279-01A1.