Dynamic and Partially Connected Ring Topologies for Evolutionary Algorithms with Structured Populations

A stigmergic presentation by Juanlu Jiménez and JJ Merelo for #evostar2014 http://evostar.org

Transcript

1. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Dynamic and Partially Connected Ring Topologies for Evolutionary Algorithms with Structured Populations A stigmergic presentation C.M. Fernandes, J.L.J. Laredo, J.J. Merelo, C. Cotta, A. Rosa 1 / 13
2. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Outline 1 Introduction 2 Dynamic, partially connected topologies 3 Reproduction 4 Results at a glance 5 Conclusions 2 / 13
3. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions What’s the problem? 3 / 13
4. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Let’s solve it! 4 / 13
5. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions A key concept: Emergence • Social animals like ants cooperate, which results in emergent behaviors • But not only ants 5 / 13
6. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions A key concept: Emergence • Social animals like ants cooperate, which results in emergent behaviors • But not only ants 5 / 13
7. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions A key concept: Emergence • Social animals like ants cooperate, which results in emergent behaviors • But not only ants Emergence [Wikipedia] Emergence is the way complex systems and patterns arise out of a multiplicity of relatively simple interactions. 5 / 13
8. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Cellular EA - Ring Topology 6 / 13
9. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Cellular EA - Ring Topology 6 / 13
10. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Cellular EA - Ring Topology 6 / 13
11. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Cellular EA - Ring Topology 6 / 13
12. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Who’s breeding? 7 / 13
13. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Who’s breeding? 7 / 13
14. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Who’s breeding? 7 / 13
15. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Who’s breeding? 7 / 13
16. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Gaps and clusters • DPCT dynamics resemble traffic jams. • Gaps and clusters form spontaneously. • n Y (Density) • Movement criterion • Phenotype (Fitness) (f) • Genotype (similarity) (s) • Random (r) • Cellular-Island Hybrid!? 8 / 13
17. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Gaps and clusters • DPCT dynamics resemble traffic jams. • Gaps and clusters form spontaneously. • n Y (Density) • Movement criterion • Phenotype (Fitness) (f) • Genotype (similarity) (s) • Random (r) • Cellular-Island Hybrid!? 8 / 13
18. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Results using trap functions Settings • n = 400 • 2-trap l = 500 • 3-trap l = 375 • 4-trap l = 300 • Top. (torus) (ring) (DPCT) 9 / 13
19. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Influence of density Settings • n = 400 • 4-trap l = 300 10 / 13
20. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions What about optimal population sizes? Settings • n=400 → optimal • 4-trap 11 / 13
21. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Conclusions • Partially connected 1-dimensional cellular GA • The resulting structure displays an island-model behaviour • Promotion of genetic diversity and reduction of the minimum population size 12 / 13
22. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Conclusions • Partially connected 1-dimensional cellular GA • The resulting structure displays an island-model behaviour • Promotion of genetic diversity and reduction of the minimum population size Future works • Extension to 2-dimensional model • Modelling the DPCT in a probability-based model 12 / 13
23. Introduction Dynamic, partially connected topologies Reproduction Results at a glance Conclusions Questions? Thanks for your attention! Follow us at http://anyself.wordpress.com @geneura & @anyselfproject 13 / 13