Abstract
Predictive coding theory suggests that target-related responses in lower brain areas are "explained away" (i.e., reduced) by feedback from higher areas. In practice, predictive coding models rely on feedback signals that must be precisely tuned to each potential target; a computationally simpler and biologically more plausible strategy could be to use spatially distributed feedback affecting both the target and its surrounding region. We tested this hypothesis by measuring luminance perception around potential targets. As in (Murray et al., 2002), we used two stimulus groups: 3D shapes and random lines created by breaking the 3D shapes at their intersections and shuffling the resulting lines. These stimuli differentially activate higher areas, resulting in different amounts of predictive feedback. In each trial, one 3D shape and one random-lines stimulus were shown simultaneously on the left and right of fixation (randomly assigned); each stimulus was presented on a gray disk, and subjects were asked to compare the luminance of these disks (report the side of the brightest disk). One disk had a fixed luminance value while the other varied around that level. We created psychometric functions for the choice probability of each disk as a function of its luminance, and compared these functions for the disk behind the 3D shape versus the one behind the random lines. Results (N=11) indicate that predictive feedback affects background luminance perception. The 3D-shapes disks were perceived brighter and the random-lines disks darker by 4.37% on average. To control for any potential attention bias, we repeated this experiment while observers were engaged in a demanding rapid serial visual presentation task. The perceptual luminance gain was still present, even though attention was engaged away from the 3D shapes and random lines. Thus, predictive feedback signals are not restricted to the predicted input, but also affect the spatial region surrounding it.
Meeting abstract presented at VSS 2014