Indirectly Encoded Fitness Predictors Coevolved with Cartesian Programs
We investigate coevolutionary Cartesian genetic programming that coevolves fitness predictors in order to diminish the number of target objective vector (TOV) evaluations, needed to obtain a satisfactory solution, to reduce the computational cost of evolution. This paper introduces the use of coevolution of fitness predictors in CGP with a new type of indirectly encoded predictors. Indirectly encoded predictors are operated using the CGP and provide a variable number of TOVs used for solution evaluation during the coevolution. It is shown in 5 symbolic regression problems that the proposed predictors are able to adapt the size of TOVs array in response to a particular training data set.
KeywordsCoevolution Cartesian genetic programming Fitness prediction
This work was supported by the Czech science foundation project 14-04197S, the Brno University of Technology project FIT-S-14-2297 and the IT4Innovations Centre of Excellence CZ.1.05/1.1.00/02.0070.
- 6.Schmidt, M.D., Lipson, H.: Co-evolving fitness predictors for accelerating and reducing evaluations. In: Riolo, R., Soule, T., Worzel, B. (eds.) Genetic Programming Theory and Practice IV. Genetic and Evolutionary Computation, vol. 5, pp. 113–130. Springer, Ann Arbor (2006)Google Scholar
- 12.Vladislavleva, K.: Toy benchmarks. In: Symbolic Regression: Function Discovery and More (2011). http://www.symbolicregression.com/?q=toyProblems