1 / 37

Evolutionary Computation and Co-evolution

Evolutionary Computation and Co-evolution. Alan Blair October 2005. Overview. Evolutionary Computation. population of individuals fitness function repeat cycle of: evaluation, selection, crossover/mutation. Bit-String Operators:. Schema “Theorem”. Implicit Parallelism

xenia
Download Presentation

Evolutionary Computation and Co-evolution

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. Evolutionary ComputationandCo-evolution • Alan Blair • October 2005

  2. Overview

  3. Evolutionary Computation • population of individuals • fitness function • repeat cycle of: • evaluation, • selection, • crossover/mutation

  4. Bit-String Operators:

  5. Schema “Theorem” • Implicit Parallelism • Fitter schemas increase their representation over time • Schemas combine like “building blocks”

  6. Evolutionary Issues: • Representations • Mutation operators • Crossover operators • Fitness functions

  7. Representations • continuous parameters (Schwefel) • Bit-strings (Holland) • genetic programs (Koza) • machine language (Schmidhuber) • NN building operators (Gruau) • genotype = phenotype itself

  8. Crossover Operators • one-point • two-point • uniform • special-purpose operators • mutation only (parthenogenesis)

  9. Fitness Functions • “deceptive” landscapes • e.g. HIFF (Watson) • local optima • premature convergence • Baldwin effect • variation over time (robustness)

  10. “Gaps” in the Fossil Record?

  11. “Gaps” in the Fossil Record?

  12. Partial Geographic Isolation

  13. Punctuated Equilibria

  14. “Gaps” in the Fossil Record? • Eldridge & Gould • partial geographic isolation • punctuated equilibria • ideas for Evolutionary Computation? • “island” models • co-evolution / artificial ecology ?

  15. Co-Evolution • competitive (leapard vs. gazelle) • co-operative (insects/flowers) • mixed co-operative/competitive (Maynard-Smith) • different genes within same genome? • “diffuse” co-evolution

  16. Sorting Networks

  17. Sorting Networks #1 (Hillis) • Evolving population of networks • converged to local optimum • final network not quite as good as hand-crafted human solution

  18. Sorting Networks #2 (Hillis) • two co-evolving populations (networks and strings) • can escape from local optima • punctuated equilibria observed • better than hand-crafted solution (Tufts, Juillé & Pollack)

  19. Co-evolutionary Paradigms • machine vs. machine (Sims) • human vs. machine (Tron) • mixed co-operative/competitive (IPD) • language games (Tonkes, Ficici) • single individual ? (Backgammon) • brain / body (Sims, Hornby, Lipson)

  20. Virtual Creatures (Sims) • Evolution • running / skipping / jumping • Co-evolution • fighting for control of a cube

  21. Tournament Structures • single species • multi-species • all vs. all • round robin • all vs. best

  22. Tron

  23. Tron • population of GP players co-evolve with population of humans over the Internet (Funes, Sklar, Juillé, Pollack)

  24. Tron Results

  25. Tron Results

  26. Iterated Prisoner’s Dilemma C D • TFT ALL-C ALL-D TFT C 3, 3 0, 5 5, 0 1, 1 D

  27. Collusion

  28. Meta-Game of Learning • Co-evolution tends to provide an opponent of appropriate ability • generally helps to escape local optima • however, can create new “mediocre stable states” (collusion)

  29. Language Games

  30. Simulated Hockey

  31. Shock Results (single player)

  32. Shock Results (one-on-one)

  33. Evolutionary Robotics • too many generations - robot may get worn out • start in simulation, refine on real robot

  34. Brain/Body Co-evolution

  35. Biped Walking • dynamically stable gait • evolved parameters • start on fast approximate simulator • refine on slow accurate simulator

  36. Future Directions • massive parallelism • modularity and evolution • credit-assignment problem • society / economic models

More Related