A machine-learning inspired paper on how adaptation to natural images shapes the top-down connections in the visual cortex. We develop a hierarchical version of a deep generative model (Top-Down VAE) to test the hypothesis that top-down connections serve hierarchical Bayesian inference. On the sidelines we find that texture sensitivity, characteristic of the secondary visual cortex emerges naturally and many properties of this texture representation are faithfully born out from learning a model of natural images.
The study was spearheaded by Feri Csikor, joined by Balázs Meszéna, and Kata Ócsai:
Csikor, F., Meszéna, B., Ócsai, K. et al. Top-down perceptual inference shaping the activity of early visual cortex. Nat Commun16, 9998 (2025). https://doi.org/10.1038/s41467-025-64967-x
CSNL
Navigation Menu