Abstract
Recent fMRI studies show that Visual Category Learning (VCL) is supported by multiple neural systems in the human brain which are presumably required for handling the wide range of demands involved in different VCL tasks. In the current study we scanned 14 healthy young adult and contrasted between their brain activities in supervised VCL tasks (with informative trial-by-trial feedback) and unsupervised VCL tasks (no feedback). For each task we used distinct novel creature-like stimuli that differed from one another across several feature-dimensions (e.g. body shape, limbs size, tail length…), but only one feature-dimension was important for determining the category membership to one of two subcategories. This design mimics the ambiguity one experiences when introduced with novel objects which can be classified in different ways. In each trial participants viewed two sequentially presented creatures and were asked to determine if they belong to the same category or not. Each learning trial was composed of three sequential events – stimuli presentation, response via key press and feedback. This enabled disassociating the neural correlates of these VCL components. We examined which cortical regions were associated with processing the feedback information using individual subject analyses. Our data show a network of brain areas that exhibit higher responses only when a participant was making an error and when informative feedback was provided. This includes the anterior cingulate, bilateral frontopolar and ventrolateral cortices (13/14 subjects), intraparietal sulcus (11/14) and cerebellum (12/14). In only about half of the participants we saw similarly higher activity associated with error trials in the ventral occipito-temporal cortex (7/14) and the caudate nucleus (8/14). This suggests that neural systems involved in error detection and conflict management (anterior cingulate), cognitive control (frontopolar; ventrolateral), visuospatial processing (intraparietal sulcus) and response learning (cerebellum) are the most critical for supporting VCL tasks by processing negative feedback.
Meeting abstract presented at VSS 2012