Infant Populations and Lateral Genetic Mobility vs. Complexification
- The beneficial objectives of Supervisor NEAT are twofold:
a) Enable implementation of pyramid number of populations vs.
starting genome complexity; and
b) Enable implementation of automated evolution of embedded
genome/populations in Embedded NEAT.
Investigation has demonstrated each of the following:
1) NEAT parameters have a profound impact on relative dominance of
lateral genetic mobility vs. complexification;
2) Standard NEAT parameter values lack sufficient lateral genetic
mobility to "get within" the genetic mobility radius of an optimal
solution in lower dimension space, prior to being weighed-down with
the associated computational overhead of increased complexity due to
3) Exploitation of infant populations to maximize lateral genetic
mobility and radius is more beneficial in identifying optimal
solution space despite higher dimensionality of NEAT parameter space
associated with Supervisor NEAT, when coupled with a pyramid
population vs. starting genome complexity scheme;
Current investigation involves the determination of (8) control
constants for multiple time histories comprising 40,000 time points,
each with associated (4) input floats and (1) target output float.
One beneficial characteristic usage of Supervisor NEAT with infant
populations, is the ability to identify infant populations
establishing new "high-water-marks" of fitness, and "getting
another" population with the same NEAT parameter set of values
(i.e. "going where the gold is").