Cec09

download Cec09

If you can't read please download the document

Transcript of Cec09

1. Lourdes AraujoUNEDJuan J. Merelo GuervsUniv. de Granada MULTIKULTIALGORITHM: USING GENOTYPIC DIFFERENCES IN ADAPTIVEDISTRIBUTED EVOLUTIONARY ALGORITHM MIGRATION POLICIES 2. Motivation

  • Island model improves sequential exec. 3. Asynchronous models better 4. Asynchronous populations better

Diversity advantages 5. Hypothesis

  • Exchanging best individuals is not always the best

Population 1 Population 2 6. Hypothesis ? ? ? ? ? ? 7. Hypothesis 8. Criteria to select migrant individuals

  • The mostdifferentto the receiving population. 9. The mostdifferentto the receiving population taking from anelite

10. Which is the most different to the whole destination population? Representing the population:

  • The best
  • Consensus sequence

11. Consensus sequence Composed of the most frequent allele for each gen.101 1 10 01 10 0 0101 10 01 1 10 0101 10 0 0101 1 0 01 1 1 10 0 0 010101 1 1 CS:0 01 1 10 0 010 0 0101 1 1 12. Our model Genotype representation G2 rather differentto G2 G1 G3 rather differentto G3 P3 P1 P2 Pn 13. Implementation

  • Chromosomes:fixed length binary strings 14. Crossover:two points and GBX (respects gene boundaries) 15. Mutation:single-bit-flip 16. Perl implementation 17. Simulated parallel environment

18. Testedproblems P-Peaks: P=100,N=64, 100, 128 19. Tested problems fitness si (0) = 1.0fitness si(1) = 0.0 fitness si(2) = 0.360384fitness si(3) = 0.640576 fitness si(4) = 0.360384fitness si(5) = 0.0 fitness si(6) = 1.0 MMDP: K=15, 20 20. Results: P-Peaks(N=64, 2 nodes) 21. Results: P-Peaks(N=64, 4 nodes) 22. Results: P-Peaks(N=64, 8 nodes) 23. Results:MMDP(90 bits, 2pX, 2 nodes) 24. Results:MMDP(90 bits, 2pX, 4 nodes) 25. Results:MMDP(120 bits,GBX, 2 nodes) 26. Entropy: 27. Entropy: averages 28. Future works Conclusions

  • Diversity improves the results provided 29. the individual is good enough 30. Entropy tests reveal a real increase in 31. diversity 32. Both, the best individual and the consensus sequence are good representatives of the population

Conclusions

  • Parallel implementation
  • Studying other parameters and problems

33. Thanks! Any question?