Complex visual scenes are made up of many component features, such as edges and textures. Neurons in early stages of the visual system are sensitive to individual features, and it is implicitly believed that the nervous system must put them back together to signal conjunctions of different features, but how this is achieved is unknown. This paper proposes a model in which neural activity encodes statistical variations of features in images, thereby allowing the visual system to generalize across variable images.
- Yan Karklin
- Michael S. Lewicki
