Artificial Cell Model Used for Information Processing

Background

In 2003, Ken Stanley and Risto Miikulainen developed a methodology to classify the different models that appear in Evolutionary Computation (EC), which have defined the new Artificial Embryogeny (AE) area. This methodology is focused on the models which are inspired in abstractions of the embryological cells. This new research area has been called by different names, like Computational Embryology or AE, by different authors. The models try to keep features such as self-organizing, self-repairing, fault tolerance and parallel information processing, which are present in the biological model, in an abstraction to apply them to different problems.

Following this classification methodology, AE works can be classified in two main types which face the problem in two different ways. On one hand, works that are included under the grammatical approach can be found. These works are related with Lindermayer’s studies and L-systems, which perform a top-down approach to the problem (1968). On the other hand, other works, with a chemical approach, are found. These last works are based in Turing’s ideas and perform a bottom-up approach to the problem (1952).

Grammatical approach works have been mostly used to develop Artificial Neural Networks (ANNs). This process is called neuroevolution. The first neuroevolutive system was developed by Kitano (1990). This work shows how the connectivity matrix of an ANN can be evolved with a set of rewriting rules. Another remarkable work is the one developed by Hornby and Pollack (2002), where the authors develop both the structure of a body in a simulated 3D world and an ANN to control it. In this case, the authors use the L-systems to develop both parts. Finally, it is necessary to mention the work developed by Gruau (1994), where the authors use a grammatical tree to store the development of an ANN from a unique starting element.