A network of neurons which are assigned onto a 2-dimensional plane represent the cells in lateral geniculate nucleus (LGN) which are the inputs to the visual cortex and are not orientation selective. Another group of neurons which are also assigned onto a 2-dimensional plane represent the cells in visual cortex (V1) which receive feedforward connections from LGN cells and also receive feedback connections from other V1 cells. The network is illustrated in Figure 1.
Figure 1: There are two types of connections for cortical cells: the feedforward connections from input LGN cells (within the rectangle connected with solid lines) and the massive feedback connections from V1 cells (within the rectangle connected with dotted lines). Due to the massive feedback among the participating cortical neurons, each cell acquires its response via a collective computation.
Each V1 cell receives feedforward connections only from the LGN cells within a specified area. Each V1 cell also receives feedback connections from other V1 cells within a small area (the size does not matter much as long as comparable to the size of the specified input area). Initially, the strengthes of the connections are small random numbers. Natural images are used as the inputs to the network. The images are first filtered with the known DOG (difference of Gaussian) kernel for LGN.