Component-Based Adaptive Architecture with Classifier Systems
Rolf Pfeifer, Bruce M. Blumberg, Jean-Arcady Meyer, Stewart W. Wilson · 1998
This paper proposes an architecture, called genetic architecture, for adaptive organisms to evolve its structures dynamically with environmental changes. This study is inspired by characteristics of animals' genetic systems that can be regarded as a set of switches to change their body structures through interactions with the environment. The genetic architecture consists of two parts; a task execution module and a structure decision module that correspond to phenotype and genotype respectively. The task execution module is responsible for tasks that the organism is faced with, whereas the structure decision module selects and produces components of the task execution module appropriately for the current situation. The task execution module employs software objects for structural components. The structure decision module consists of a classifier system to learn associations between situations and organisms' structures. We conducted some experiments, and results show validity of the genetic architecture.