Basic Principles of Adaptive Learning through Variation and Selection
Mikhail Burtsev · Artificial Life · 2008
The evolutionary theory relies on the principles of variation and selection to explain adaptation. It is reasonable to fit these powerful principles to the learning theory. A number of selectionist approaches were proposed but found modest recognition so far. This theoretical paper attempts to review an application of basic ideas of the evolutionary adaptation to the lifetime learning. The analysis demonstrates that an adaptive value can be translated from the level of evolution to the level of individual through the innate repertoire of behaviors. This primary repertoire forms initial attractor for the behavioral dynamics. Learning starts when an environment offsets an organism from the existing attractor trajectory. Blind variations of behavior are generated until the return to the target attractor. These variations are retained to make up new branches of basin of attraction. It is important that the existed behavioral trajectory should not be altered as the learning unfolds because it keeps knowledge about adaptations survived selection through the evolutionary and learning history.