Genetic programming: on the programming of computers by means of natural selection
Genetic programming: on the programming of computers by means of natural selection
Genetic programming: an introduction: on the automatic evolution of computer programs and its applications
Genetic Programming for Classification: An Analysis of Convergence Behaviour
AI '02 Proceedings of the 15th Australian Joint Conference on Artificial Intelligence: Advances in Artificial Intelligence
Experimental Study of Multipopulation Parallel Genetic Programming
Proceedings of the European Conference on Genetic Programming
Program simplification in genetic programming for object classification
KES'05 Proceedings of the 9th international conference on Knowledge-Based Intelligent Information and Engineering Systems - Volume Part III
Hi-index | 0.00 |
The genetic programming (GP) search method can often vary greatly in the quality of solution derived from one run to the next. As a result, it is often the case that a number of runs must be performed to ensure that an effective solution is found. This paper introduces several methods which attempt to better utilise the computational resources spent on performing a number of independent GP runs. Termed meta-search strategies, these methods seek to search the space of evolving GP populations in an attempt to focus computational resources on those populations which are most likely to yield competitive solutions. Two meta-search strategies are introduced and evaluated over a set of classification problems. The meta-search strategies are termed a pyramid search strategy and a population beam search strategy. Additional to these methods, a combined approach using properties of both the pyramid and population beam search methods is evaluated. Over a set of five classification problems, results show that meta-search strategies can substantially improve the accuracy of solutions over those derived by a set of independent GP runs. In particular the combined approach is demonstrated to give more accurate classification performance whilst requiring less time to train than a set of independent GP runs, making this method a promising approach for problems for which multiple GP runs must be performed to ensure a quality solution.