Information Technology Reference
In-Depth Information
Table 10.1
Parameters for the XOR problem using a redundant and a compact system.
Redundant
Compact
Number of runs
100
100
Number of generations
50
50
Population size
30
30
Number of fitness cases
4
4
Function set
D T Q
D T Q
Terminal set
a b
a b
Weights array length
10
10
Weights range
[-2, 2]
[-2, 2]
Head length
4
2
Number of genes
1
1
Chromosome length
33
17
Mutation rate
0.061
0.118
Inversion rate
0.1
--
One-point recombination rate
0.3
0.3
Two-point recombination rate
0.3
0.3
IS transposition rate
0.1
--
RIS transposition rate
0.1
--
Dw mutation rate
0.061
0.118
Dw-specific inversion rate
0.1
0.1
Dw-specific transposition rate
0.1
0.1
Weights mutation rate
0.01
0.01
Fitness function
Eq. (3.8)
Eq. (3.8)
Success rate
77%
30%
problem, was to show the plasticity of GEP neural networks and that they
also thrive in slightly redundant architectures (see, for instance, a discussion
of The Role of Neutrality in Evolution in chapter 12). And as you can see in
Table 10.1, the success rate for this problem using the redundant chromosomal
organization is higher (77%) than the obtained with more compact organizations
with
h
= 2 (30%). And it is thanks to this plasticity that the neural network
architecture can evolve without any kind of human intervention, contrasting
sharply with conventional neural networks where the structure is chosen a
priori for each problem and maintained unchanged throughout the learning
process.
On the other hand, as you already know, gene expression programming
can be useful for searching parsimonious solutions, and very interesting
