Learning in the AMS Context

Abstract

In this chapter, we dig further into the notion of “learning” within the AMS context. In conventional connectionist models, the term “learning” is almost always referred to as merely establishing the input-output relations via the parametric changes within such models, and the parameter tuning is typically performed by a certain iterative algorithm, given a finite (and mostly static) set of variables (i.e. both the training patterns and target signals). However, this interpretation is rather microscopic and hence still quite distant from the general notion of learning, since it only ends up with such parameter tuning, without giving any clear notions or clues to describe it at a macroscopic level, e.g. to explain the higher-order functions/phenomena occurring within the brain (see e.g. Roy, 2000).