Abstract
Information bottleneck limits feedforward signals from the primary visual cortex (V1) to higher brain areas, these signals nevertheless give initial perceptual hypotheses about visual scenes. Feedback from higher to lower visual areas verify and re-weight these hypotheses in noisy or ambiguous situations via analysis-by-synthesis. We call this process Feedforward-Feedback-Verify-reWeight (FFVW). CPD hypothesizes that the feedback is weaker or absent in the peripheral visual field (Zhaoping, 2017, 2019). Accordingly, peripheral vision is vulnerable to illusions by misleading V1 signals. CPD and FFVW are manifest using RDSs for depth surfaces by contrast-reversed random-dots (CRRDs), for which a black dot in one eye corresponds to a white dot in the other eye. V1 neurons respond to CRRDs as if their preferred binocular disparities become anti-preferred and vice versa, signalling reversed depths. I show that (1) the reversed depth is seen in peripheral but not central vision; (2) compromising feedback by brief viewing and backward masking makes reversed-depth seen in central vision; and (3) adding reversed-depth signals to normal-depth ones enhances or degrades depth percepts in central vision when these depth signals by V1 responses agree or disagree, respectively, with each other. In (3), degradation occurs only when the feedback is compromised, whereas enhancement occurs regardless, revealing a nonlinearity in the feedback processes for perceptual decisions.
Funding: University of Tübingen, Max Planck Society