Abstract
Learning to group diverse visual stimuli into categories is correlated neurally with increased within-category pattern similarity and increased between-category pattern separation in high-level visual cortex (Folstein et al. 2015; Clarke et al. 2016; Hammer & Sloutsky 2016). However, the causal link between the neural representation of categories and their perception remains unclear. To address this question, we use closed-loop real-time fMRI neurofeedback to induce neural plasticity (Jackson-Hanen et al. 2014; deBettencourt et al. 2015), similar to what we would observe during category learning, but underneath the threshold of awareness and without any explicit learning taking place from the participants. More specifically, we constructed a stimulus space of complex artificial shapes that varied along multiple dimensions independently and confirmed (n=750) that each stimulus dimension is perceived in an equivalently graded manner, as are manipulations of the space along these dimensions. Moreover, multiple brain regions represented this space as a putative cognitive map, mirroring perception (EVC, LOC, PFC, temporal pole). Within this space, we hypothesized that using neurofeedback to strengthen unique features and suppress shared features of visual categories should also differentiate the categories perceptually. To compute the neurofeedback provided to differentiate categories, we developed a novel computational approach (KL-Evidence) based on mutual information between the distributions of neural responses they elicit. We will present preliminary evidence that this procedure influences perceptual similarity of the items after training, that is, that we induced the presence of stronger implicit perceptual categories along the trained dimension, compared to the untrained dimensions. This suggests a potential causal link between neural representations in these brain regions and perception. More broadly, this technique shows promise for becoming a novel conduit to access and manipulate the contents of complex subjective visual experience, as well as learning in a non-invasive way via fMRI.
Acknowledgement: John Templeton Foundation, Intel Corporation, NIH R01 MH069456