OnMapping
The general state space of our system under scrutiny is too large to think about: each recognizable attribute is a dimension, making a multi-dimensional space to roam through. Rarely (ever?) does the curiousity actually inhabit the whole space. Instead, it occupies a chunk of the space we can view as a manifold (object) occupying the multi-dimensional. The manifold may be continuous or discontinous, big or small, dynamic or static.
An attempt to map a curiousity is an attempt to discover the manifold. We play the game of finding some sort of low-resolution, coarse representation of the manifold of intrigue; an attempt to widdle our large state space down to something more manageable, something more available to exploration. Ideally, a point on the manifold will map back to an entity bearing resemblance to other instances from our system under inquiry.
A deficiency of many mapping methods is the manifold is created from current knowledge with ways to prune its coarse shape, but no process for new growth. This deficiency can be a feature if we wish to keep a strict bound on the manifold. Strict bounds are nice when we want to model known qualities. When we want to derive novelty from known data, we want our manifold to bloom. GAs1 can (always?) suffer from this defficiency.
1 Genetic Algorithms
Last Edit: Sun, 23 May 2004 18:56:57 -0700 Revisions: 5