Researchgenerative modelsvisual cortexpredictive codingself supervised learning
Mouse Visual Cortex Predicts Higher-Level Features
8.3
Relevance ScoreHeilbron and de Lange (published January 20, 2026) used deep generative modeling to quantify spatial predictability of receptive-field image patches and compared these estimates to high-density recordings from the Allen Institute Visual Coding Dataset across mouse visual cortex. They report that more predictable patches evoke weaker responses, sensitivity is strongest for higher-level features (even in V1), is elevated in superficial layers, and reflects long-term priors.
