S2 Layer

The S2 layer is the second “selectivity” cell layer. It computes a weighted sum over a set of C1 inputs. All S2 units are alike, except that they differ in the locations of the receptive fields (i.e., the position of the fields of S1 units that their C1 units pool over). The parameters controlling the S2 function are set by the <s2> object:

<s2>
  sigmoid_s       1.127482
  sigmoid_alpha   1.925626
  sigmoid_beta    0.596638

#
#   x goes from 0 to (xn-1) (specified in C1 grid
#
# grid  x y ori     w
# ----- - - -    ---------
  c1_48_1_1_2    -0.865299
  c1_48_2_1_0     1.148643
</s2>

The first three parameters set the amplitude (s), slope (alpha) and horizontal offset (beta) of a sigmoid function. The remaining parmeters specify weights for C1 units, where the parameter name specifies the C1 grid name, the indices of the x,y location within the grid, and the index within the set of orientations.