TABLE 1.

Model parameters

SCALES1 RFC1 RFC1 SHIFTC1 GRIDS2 RFS2 SHIFTS2 GRIDC2 RF
154, 6080402×2
240, 4560303×3120303×3180
332, 3648244×4
  • The S1 and C1 layers are broken down into three spatial scales (1st column). The receptive field (RF) sizes of S1 units vary across spatial scales (2nd column, measured in pixels). Within each spatial scale, CI units receive input from S1 units with 2 different RF sizes (to achieve a small degree of scale invariance) and with different spatial locations (to achieve translation and phase invariance). The resulting RF sizes of CI units are indicated in the 3rd column. Within each spatial scale, CI units form spatial grids with the center of adjacent CI receptive fields shifted by the amount indicated in the 4th column. In the S2 layer, S2 units receive input from all 3 spatial scales. The CI grid sizes for each spatial scale span the same range of space (120×120 pixels), giving S2 units an identical RF size. S2 units form a 3×3 grid with adjacent S2 units shifted by 30 pixels. The top layer C2 unit has a RF size of 180 pixels. In the model, 32 pixels correspond to ∼1° of visual angle.