ABSTRACT
In other words, A†A maps some components of y to , but maps others to 0, depending on the size of each eigenvector. This eliminates the smallest eigenvectors of A and regularizes the solution.
Compare this with what we derived in §0.3. Here, the eigenvalues of A were shifted so that ?i? ?i,+s N
2. This transforms the problem so that no eigenvalue of A
is less than s N 2, thus regularizing the problem in another way. In principle, these
two approaches should lead to similar results.
