Top-down generation of low-resolution representations improves visual perception and imagination

Abstract

According to analysis-by-synthesis theories of perception, the primary visual cortex (V1) reconstructs visual stimuli through top-down pathway, and higher-order cortex reconstructs V1 activity. Experiments also found that neural representations are generated in a top-down cascade during visual imagination. What code does V1 provide higher-order cortex to reconstruct or simulate to improve perception or imaginative creativity? What unsupervised learning principles shape V1 for reconstructing stimuli so that V1 activity eigenspectrum is power-law with close-to-1 exponent? Using computational models, we reveal that reconstructing the activities of V1 complex cells facilitate higher-order cortex to form representations smooth to shape morphing of stimuli, improving perception and creativity. Power-law eigenspectrum with close-to-1 exponent results from the constraints of sparseness and temporal slowness when V1 is reconstructing stimuli, at a sparseness strength that best whitens V1 code and makes the exponent most insensitive to slowness strength. Our results provide fresh insights into V1 computation.

The attentive reconstruction of objects facilitates robust object recognition

“…One question not answered by our current work is how completely an object must be reconstructed in order to serve as an effective attentional template. Our finding that even low-resolution object reconstructions are sufficient to generate attentional biases and performance benefits align with previous studies showing that the brain initially processes low spatial frequency information about objects for the purpose of generating top-down feedback signals for guiding the bottom-up processing of finer-grained details (Bar, 2003; Bar et al, 2006; Bi, 2021). However, effects of feature-based attention are also found in a range of specific low-level features (e.g., color, orientation, etc, Maunsell & Treue, 2006), making further computational experiments needed to determine whether optimization of the information and level of object reconstruction might yield even greater performance benefits.…”

Predictions enable top-down pattern separation in the macaque face-processing hierarchy

“…Our results so far suggest that lower areas inherit tuning properties from higher areas through predictions. Because according to predictive processing theory, PEs are computed relative to predictions generated in higher areas, we hypothesized that PEs in lower areas should also express higher-level tuning properties 7,[22][23][24] . Alternatively, PEs could enhance feedforward information, i.e., the local tuning properties of the area in which they are generated 25 .…”