Chris Watkins
ABSTRACT
We present a population genetic algorithm which satisfies detailed balance, and which has a stationary distribution that factorises into an explicit form for arbitrary fitness functions. For a population size of 1, it is the Metropolis algorithm with a `breeder' proposal distribution; it extends to larger populations in a natural way, and the stationary (that is, the mutation-selection equilibrium) distribution is exactly known in a simple form for any population size. We term this algorithm exchangeable breeding tuple product sampling (EBT).
EBT is closely related to some non-parametric Bayesian Markov-chain Monte Carlo sampling algorithms. EBT can also be viewed as a generalisation of the Moran process.
- D. Goldberg. Genetic algorithms in search, optimization, and machine learning. Addison-wesley, 1989. Google Scholar
Digital Library
- J. Holland. Adaptation in natural and artificial systems. University of Michigan press, 1975.Google ScholarDigital Library
- O. Kallenberg. Probabilistic symmetries and invariance principles. Springer Science & Business Media, 2006.Google Scholar
- J. G. Kemeny and J. L. Snell. Finite Markov chains: with a new appendix" Generalization of a fundamental matrix". Springer, 1976.Google Scholar
- R. Neal. Markov chain sampling methods for dirichlet process mixture models. Journal of computational and graphical statistics, pages 249--265, 2000.Google Scholar
- P. Orbanz and Y. W. Teh. Bayesian nonparametric models. In Encyclopedia of Machine Learning, pages 81--89. Springer, 2010.Google Scholar
- C. J. Ter Braak. A markov chain monte carlo version of the genetic algorithm differential evolution: easy bayesian computing for real parameter spaces. Statistics and Computing, 16(3):239--249, 2006. Google ScholarDigital Library
- M. Vose. The simple genetic algorithm: foundations and theory. The MIT Press, 1999. Google ScholarDigital Library
Index Terms
- Efficient Sampling with Small Populations: a Genetic Algorithm Satisfying Detailed Balance
Recommendations
Dynamic populations in genetic algorithms
SAC '08: Proceedings of the 2008 ACM symposium on Applied computingBiological populations are dynamic in both space and time, that is, the population size of a species fluctuates across their habitats over time. There are rarely any static or fixed size populations in nature. In evolutionary computation (EC), ...
Probabilistic model-building genetic algorithms
GECCO '08: Proceedings of the 10th annual conference companion on Genetic and evolutionary computationProbabilistic model-building algorithms (PMBGAs) replace traditional variation of genetic and evolutionary algorithms by (1) building a probabilistic model of promising solutions and (2) sampling the built model to generate new candidate solutions. ...
Probabilistic model-building genetic algorithms
GECCO '10: Proceedings of the 12th annual conference companion on Genetic and evolutionary computationProbabilistic model-building algorithms (PMBGAs) replace traditional variation of genetic and evolutionary algorithms by (1) building a probabilistic model of promising solutions and (2) sampling the built model to generate new candidate solutions. ...
Comments