The talk focuses on learning algorithms inspired, in particular,
by Tikhonov regularization. We discuss some mathematical
properties of the underlying functionals including the
problem of parameter estimation viewed as the key to
obtain effective generalization.