Seznam |

Téma: | Restarting, or meta-optimization? |
---|---|

Vedoucí: | Ing. Petr Pošík Ph.D. |

Vypsáno jako: | Diplomová práce |

Popis: | Local search (LS) algorithms find only local optima of optimization problems. Their chance to find the global optimum is usually increased by restarting them. There are, however, many strategies when and how to restart the algorithm; many of them are parameterized. Finding the optimal strategy is a difficult problems in itself. Meta-optimization algorithms constitute an alternative to the restarting. These algorithms contain several instances of LS algorithms running side-by-side. In each iteration, they decide which of the portfolio members get the chance to make another step. Restarting is replaced by running many LS instances in various areas of the search space; the decision when to restart an algorithm is thus replaced by the decision which algorithms shall make another step. The goal of this project is to compare the restarting strategies and meta-optimization algorithms regarding the complexity of finding the optimal parameter values and regarding the performance of the resulting algorithm |

Pokyny: | 1) Perform a survey of existing restarting strategies and meta-optimization algorithms and try to categorize them.
2) Select at least 2 universal restarting strategies and at least 2 metaoptimization algorithms, find their optimal settings for a chosen set of reference optimization problems and compare them. 3) Perform a statistical analysis of the results. |

Literatura: | Will be provided by the supervisor. |

Realizace: | implementation, experiment results, final report. |

Vypsáno dne: | 07.02.2013 |