On Mon, 15 Oct 2001, Oleg Golubitsky wrote:
> ................................................................
> 3. The search for a subclass that best fits the training set is
> defined as a search for an element in the power class (Section 4), and
> thus, in principle, is not much different from the search for an
> element in the class or the computation of an element's
> typicality. However, a formal definition of the power class is needed to
> completely clarify this analogy. We believe that there is a tight connection
> between the descriptions (transformation systems, Def. 3) of the class and its
"Def.3" should read "Def. 30"
> corresponding power class; we are currently working on this definition.
> When developed, it will also clarify what the learning complexity is,
> since the latter depends on the power class description.
