Characterizing receptive field selectivity in area V2

Model modulation index Figure 1 (a) An input image is convolved with a model V1 front-end with units tuned to different orientations and spatial frequencies. The first V2 stage computes derivatives over the activation map of V1 across space, as well as orientation and spatial frequency. We modelled “V2 simple cells” as the rectified output of these filters. The final V2 stage pools and rectifies the output of these “V2 simple cells”. (b,c) Single unit modulation indices for V1 and V2 cells adapted from Freeman et al. (2013). The modulation index represents the normalized difference in firing rate between naturalistic and control stimuli. The V1 neurons show an average modulation close to zero, whereas the V2 population was consistently positive. (d,e) Modulation indices for a simulated population of V2 “simple” and “complex” cells. Units were simulated to have random parameters for derivative and pooling weights. While “V2 simple” cells had a slightly positive modulation index on average, “V2 complex” cells approached the magnitude of the recorded V2 population. (f) The average modulation of “V2 simple” cells failed to account for the strength of V2 modulation across different categories of naturalistic images, (g) while The average modulation of “V2 complex” cells across categories was correlated with recorded V2 population. Characterizing receptive field selectivity in area V2 Corey M Ziemba, Robbe LT Goris, J Anthony Movshon, Eero P Simoncelli New York University