Parameter Estimation for Linear Models

In this chapter we discuss various generalizations of the ubiquitous linear regression problem which appear in many data modeling circumstances. We discuss multivariate models from the outset. The technique to solve the problem turns essentially out to be just least squares which, for Gaussian data, can be directly justified based on the maximum likelihood principle. We warn the reader that this is however true only if it is based on strong a priori assumption of noiseless output data and suggest a wide perspective.