CGP is a form of genetic programming that represents computer programs as directed acyclic graphs encoded as strings of integers. Programs are evolved using genetic operations like mutation and crossover. Mutation changes integer values representing connections, functions, inputs, and outputs. Crossover combines parts of two genotype strings. The genotype is decoded into a phenotype program. CGP uses strategies like 1+λ evolution and can experience genetic redundancy from unused genes or sub-functions not optimized.

1. J.F. Miller
Jagdeep Matharu - 4831400
Seminar 4V82

2. What is CGP?
 CGP is a form of Automatic computer program
Evaluation (GP)
 Developed be Miller and Thompson 1997.
 Inspired from evaluation of digital circuit.
 Capable of encoding computer programs, electronic
circuits, neural network.

3. Representation
 Programs are represented as directed acyclic graphs
which are encoded in the form of a linear string of
integer
 Genes are
 Address in data (Connection genes)
 Address in a function lookup table (Function genes)
 Address in output data (Output genes)
 Genotype is string of integers.
Eg. 0 0 1 1 0 0 1 3 1 2 0 1 0 4 4 2 5 4 2 5 7 3

4. CGP Genotype

5. CGP General form

6. Cont’d

7. Genotype-to-Phenotype mapping
 Result from the decoding of a genotype is called
phenotype.
 Many-to-one genotype to phenotype mapping.
 Some genes in phenotype can be ignored

8. Decoding
Genotype
Phenotype

9. Evolution of CGP Genotypes
 Most CGP system use only mutation.
 Point-mutation
 Mutation rate
 Gene location is change with other valid random value.
 Function with other random valid address of function.
 Input gene value with valid output from any other node
or terminal node value.
 Output with address of output of other node in
genotype or terminal node value.
 Crossover

10. Cont’d

11. Evaluation strategies
 1+𝜆 algorithm

12. Cont’d
 An offspring is always chosen if it is equal as fit or has
better fitness than the parent.

13. Genetic Redundancy
 Node redundancy
 Genes those are not used in fitness calculation.
 Functional redundancy
 Sub-function that actually may be implemented with fewer
nodes
 bloat
 Input redundancy
 Node functions are not connected to some of the input node
 Neutrality
 Adaptive evolution may cross regions with poor fitness in
fitness landscape.

14. References
“CGP Home.” Accessed November 27, 2012.
http://www.cartesiangp.co.uk/
J.F. Miller(ed.), Cartesian Genetic Programming ,
Natural Computing Series, DOI 10.1007/978-3-642-17310-
3 2,