In applying our theory of associative decorrelation dynamics to visual cortex to compare with the psychophysical experiments on orientation illusions, the linear first-order approximation is used, which is
where it is assumed that the input correlations are small. It is interesting to notice that the linear first-order approximation leads to anti-Hebbian feedback connections: which is guarantteed to be stable around .
Let us to illustrate the consequences of this for a simple two neurons case. Suppose the , that is the original inputs ensemble is uncorrelated. Now the network is adapted to an input , what is the network response to an input now? From equation 17
So the response after adaptation is actually rotated from the ``vertical'' of input , the angle rotated is given by