Designing an Estimation of Distribution Algorithm based on Learning Automata

Moradabadi, Behnaz | 2011

434 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 42878 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Beigy, Hamid
  7. Abstract:
  8. Evolutionary algorithms are a type of stochastic optimization techniques influenced by genetics and natural evolution. Once the set of candidate solutions has been selected, a new generation is sampled by using recombination (crossover) and mutation operators to the candidate solutions. Public, fixed, problem independent mutation and recombination operators frequently lead to missing building blocks, knowledge of the relationship between variables and result in converging to a local optimum. A method to prevent disruption of building blocks is using the estimation of distribution algorithms (EDAs). The experimental results show that EDAs is capable to identify correct linkage between the variables of an optimization problem. But finding the optimal probability distribution model is an NP-hard problem. Also learning automata (LA) is an adaptive decision making unit that tries to learn the optimal action from an action-set by interacting with a random environment. In each step, it selects an action from its action-set. The action selection is based on a probability distribution over the action-set. The selected action is applied on the environment and a reinforcement signal is produced by the environment. LA updates its actions probability distribution according to the produced reinforcement signal and a learning algorithm. Then it chooses an action again and these steps are repeated until LA converges to some action. This thesis proposes some methos to improve the efficiently of EDAs. It applies LAs to propose some EDAs. The experimentl results show the supreiority of the proposed algorithms
  9. Keywords:
  10. Evolutionary Algorithm ; Bayesian Network ; Learning Automata ; Distribution Estimation Algorithm

 Digital Object List

  • محتواي پايان نامه
  •   view