Minor-components analysis (MCA) is the opposite of principal-components analysis. In MCA, we seek to find those directions that minimize the projection variance. The directions so found are the eigenvectors corresponding to the smallest (minimum) eigenvalues of the correlation matrix R of the input random vector X(n).
In this problem, we explore how to modify the single neuron of Section 8.4 so as to find the minor component of R. In particular, we make a change of sign in the learning rule of Eq. (8.40), obtaining the following (Xu et al., 1992):
