Conference on Cognitive Computational Neuroscience

Mental imagery is often described as a reactivation of sensory activity in the brain. However, recent research focusing on vision and imagery suggests that the trial-averaged voxel activity for mental images (i.e. the imagery signal) in the visual cortex varies far less compared to when the images are seen. What might explain this substantial reduction in signal variance during imagery? We consider two of the possible hypotheses: (1) the ``weak vision" hypothesis, in which the imagery signal is simply a scaled-down version of the visual signal and (2) the ``compressed vision" hypothesis, in which imagery signal is a projection of the visual signal onto a lower-dimensional subspace. To compare these two hypotheses, we use voxel-to-voxel predictive models on data from a 7T fMRI experiment in which participants viewed and imagined 12 different stimuli. Voxel-wise prediction of imagery trials seems to particularly benefit from using correlated multi-voxel patterns of activity during vision over voxel-specific responses. Accuracy can be further enhanced by assuming imagery to be a lower-dimensional projection of visual responses. These results offer provisional support for the ``compressed vision" hypothesis.