In many cases, the data may depend non-linearly on a
parameter. If the feature extraction (signal processing)
involves estimating this parameter, it may require
an iterative algorithm. A large class of these problems can be
expressed as ML parameter estimation where the
output features are ML parameter estimates.
PDF projection for problems of this type
has been discussed in Section 2.3.5.
We now discuss a concrete example in which the
data model follows a Gaussian distribution with non-linear
parameter dependence.
Subsections
