|Topic:||Fast learning in Bayesian optimization algorithm|
|Supervisor:||Ing. Petr Pošík Ph.D.|
|Description:||Algorithms ECGA (extended compact genetic algorithm) and BOA (Bayesian optimization algorithm) are population-based optimization algorithms. They are among the most powerfull methods for optimization of complex black-box optimization problems with binary representation. Each generation they build a model of the structure of dependencies among individual solution components. The model learning is a time consuming operation. For ECGA, an efficiency enhancement was proposed recently that allows to simplify and accelerate the learning without any negative effect on the algorithm performance. The goal of this project is to implement a similar method of model learning for algorithm BOA, and evaluate the potential positive and negative effects on the algorithm performance.|
|Instruction:||1) Learn the principles of algorithms ECGA and BOA.
2) Explore the method used in ECGA to accelerate the model learning.
3) Apply the method to algorithm BOA adequately.
4) On a set of benchmark problems, compare the original and the modified algorithms with respect to the number of objective function evaluations required to find the solution, and with respect to the time required to run the algorithm.
|Bibliography:|| Duque, Thyago S.P.C.; Goldberg, David E.; Sastry, Kumara: Enhancing the Efficiency of the ECGA. PPSN 2008, Dortmund.
 Pelikan, M. Hierarchical Bayesian Optimization Algorithm
|Realization form:||Modified BOA algorithm, experiment results, final report.|