1 / 23

Recurrent Genetic Algorithm: Sustaining Evolvability

Recurrent Genetic Algorithm: Sustaining Evolvability. Adnan Fakeih Ahmed Kattan. Evo * 2012 Adnan Fakeih , Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability. Evolvability.

jaime-boyle
Download Presentation

Recurrent Genetic Algorithm: Sustaining Evolvability

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Recurrent Genetic Algorithm: Sustaining Evolvability Adnan Fakeih Ahmed Kattan Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  2. Evolvability • The notion of “evolvability” is defined as “the ability of a population to produce variants fitter than any yet existing” [1] 1. Altenberg, L.: The evolution of evolvability in genetic programming. In: Kinnear Jr., K.E. (ed.) Advances in Genetic Programming, ch.3, pp. 47–74. MIT Press (1994) Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  3. Evolvability • The term evolvabilitydoes not only refer to how often offspring are fitter than their parents but also to the entire distribution of fitness values among offspring produced by a group of parents Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  4. Our Work • “survival of the fittest”. • We consider another way of looking at the term “fittest” • Individuals who are able to produce fitter offspring. Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  5. Recurrent Genetic Algorithm • RNNs can use their internal memory to process arbitrary sequences of inputs. • GA applications, in general, do not have temporal dimension as the basic focus lies in finding the best solution in a static solution space • We use the recurrencey concept in the intermediate population to capture the temporal effects that the GA undergoes during the process of evolution. Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  6. Recurrent Genetic Algorithm Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  7. Calculate Fitness • Preliminary experiments show that an evaluation of individuals that is solely based on the fitness value of their offspring may not appropriately enhance the evolvability. Why? • Several factors controlling the creation of any offspring; • such as crossover/mutation point, original fitness value of the parent, selection pressure and the fitness value of the other parent in case of crossover. Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  8. Calculate Fitness ____________________________________________________ Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  9. Calculate Fitness Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  10. Elitism • We introduced Elitism from Intermediate population to population t+1. Why? • In crossover operator, the evaluation of fitness values is dependent on the amount of chromosomes they passed to their offspring. • On one hand, it is a strength in that the parent who passed more chromosomes into its offspring is more likely to pass a good building-block. • On the other hand, it is a weakness because if this good building-block of chromosomes is a mixture of both parents (e.g., tail of the first parent concatenated with the head of the second parent). Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  11. Experiments • NK-Landscape (unimodal problem) • Hamming Centres(multimodal problem) • Three different continuous optimisationproblems. Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  12. NK-Landscape Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  13. NK-Landscape Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  14. N=20

  15. N = 30

  16. N =40

  17. Hamming Centres Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  18. Continuous Optimisation Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  19. Modify the FRF Function • Consider the values contained in the chromosomes contributed by the parents rather than the mere number of chromosomes • Thus, Contribution of Parent X now is the sum of the values contained in the chromosomes contributed into the offspring. Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  20. Continuous Optimisation Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  21. Conclusion • This paper proposes a new paradigm, referred to as Recurrent Genetic Algorithms (RGA), that attempts to sustain Genetic Algorithm (GA) evolvability and effectively improves its ability to find superior solutions. • As an experimental validation of the new paradigm on a non-trivial space and structured representation. • Experimental evidence shows that RGA remarkably maintains higher diversity and increase the population’s ability to produce fitter offspring Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  22. Future Work • RGP (Genetic Programming) • Enhance FRF functions. • Test RGA on Real-World Applications. Evo* 2012 Adnan Fakeih, Ahmed Kattan, Recurrent Genetic Algorithms: Sustaining Evolvability

  23. Thank you for paying attention!

More Related