Post on 16-Apr-2017
A Peer-to-Peer Approach to Genetic Programming
Juan Luis Jimnez Laredo, Daniel Lombraa Gonzlez, Francisco Fernndez de Vega, Maribel Garca Arenas, Juan Julin Merelo Guervs
University of Granada (1,4,5)University of Extremadura (2,3)
http://geneura.wordpress.com http://twitter.com/geneura
Picture by Taivasalla http://www.flickr.com/photos/taivasalla/2306087602/
Peer to Peer in 2 minutes
Why P2P?
Gather more computing nodes.
Avoid single point of failure.
Test on more natural computing substrates.
Picture from Jason Samfield, http://www.flickr.com/photos/jason-samfield/5246071784/
Why not P2P?
High churn.
Unknown topology.
Wide performance ranges.
Picture The great Escape, by TanakaWho http://www.flickr.com/photos/28481088@N00/3817317092/
Can we adapt genetic programming to a P2P environment?
Short answer:
Yes
But here's the long answer
Using the Evolvable Agent model:
J.L.J. Laredo, P.A. Castillo, A.M. Mora, and J.J. Merelo. Exploring population structures for locally concurrent and massively parallel evolutionary algorithms. In IEEE Congress on Evolutionary Computation (CEC2008), WCCI2008 Proceedings, pages 26102617. IEEE Press, Hong Kong, June 2008.
http://lugl.info/xiQP
Meet the (evolvable) agent
Spatially structured Evolutionary Algorithm
Every agent schedules a single individual
Neighborhood self-organized via newscast protocolMore on this later
Basic implementation: Agent == Thread
Newscast: Gossip protocol
Joining: connecting to an existing node.Leaving: stop communication
Emergent small world graph structure
Fault toleranceGraceful degradation
Picture taken from http://www.flickr.com/photos/picsoflife/4696497459/lightbox/
Newscast: Cache size 4
4312
12334
12341234
Picture taken from http://www.flickr.com/photos/picsoflife/4696497459/lightbox/
Experimental setup
Simulated P2P environment
Well known GP problems11 bit multiplexer: 11M
Even parity 5: EP5
Source available at:
http://forja.rediris.es/svn/geneura/peerevogen
Check paper for GP parameters
Results
Concluding
Parallel execution of GP using EvAg implies improvement in times to solution.
Thanks for your attention.
Any question?
(Check/download presentation at http://geneura.wordpress.com)
Pulse para editar los formatos del texto del esquemaSegundo nivel del esquemaTercer nivel del esquemaCuarto nivel del esquemaQuinto nivel del esquemaSexto nivel del esquemaSptimo nivel del esquemaOctavo nivel del esquemaNoveno nivel del esquema