Towards the Automatic Programming of NEPs - Foundations on Natural and Artificial Computation - page 304

Fig. 1. Blocks of a general way to program natural computers

- The fitness function , which must fulfill two roles:

1. Simulate the generated solution (in this case, a particular NEP).
2. Measure how well the solution solves the target problem.

This paper shows the skeleton of our implementation of this methodology for
NEPs. To generate a correct initial population is the first mandatory step to test
its viability.
In the following sections, we will briefly introduce the main topics related to
our research (AGE/CGE, NEPs). We will next describe our platform. Finally
we will highlight our further research lines.

2 Introduction to AGE/CGE

Attribute grammars (AG) [7] are one of the tools used to completely describe
high level programming languages (both their syntax and their semantics). Chris-
tiansen Grammars (CG) [3] are an adaptable extension to AG, that is, they are
attribute grammars that modify themselves while they are used.
AGE [5] and CGE [11] are extensions to Grammatical Evolution [10]. Both
techniques are automatic programming evolutionary algorithms independent of
the target programming language, and include a standard representation of geno-
types as strings of integers (codons), and a formal grammar (respectively at-
tribute and Christiansen grammar) as inputs for the deterministic mapping of a
genotype into a phenotype. This mapping minimizes the generation of syntac-
tically and also semantically invalid phenotypes. Genetic operators act at the
genotype level, while the fitness function is evaluated on the phenotypes.
Further details, deeper descriptions and examples can be found in [5] and [11].