poster js eo-300dpi
TRANSCRIPT
An object-oriented library in JavaScript to build modular and flexible cross-platform evolutionary algorithms
Department of Computer SciencesUniversity of Jaen (Spain)
Department of Computer's Architecture and TechnolgyUniversity of Granada (Spain)
Victor M. Rivas Juan J. Merelo, Gustavo Romero, Maribel García Arenas, Antonio Mora García
The GeNeura Team's Evolutionary Object Libraries Family
1990's EO C++
2000'sJEO Java
OPEAL Perl
C++
2010'sjsEO
Javascriptnodeo
“Any object that can be attached some kind of fitness value is a potential candidate for evolution.”
github.com/vrivas/jsEO
√ Portability: 99,99% Navigator+OS
√ Trivially “installed” and executed
x Slow
x Low amount of computer resources
<<abstract>>jsEO
<<abstract>>jsEOIndividual
jsEOPopulation
<<abstract>>jsEOAlgorithm
<<abstract>>jsEOOperator
Makes to evolve
Uses
Is composed of
jsEO's Class Diagram: Core package
jsEOGetIndividual
jsEOSendIndividual
Distributed island model
X
Server
EAs are executed in many different navigators that run under many differents operating systems
installed in many different hardware devices
Best LOCAL individual
Best GLOBAL individual
Experiments and Results
256-bit Royal Road function128-real terms equation
Executed just accesing:jseo.vrivas.es
Two problems
Population size: 500Tournament size: 2Generations: 50Crossover rate: 0.73Mutation rate: 0.18Requesting individual rate: 0.09Genes affected by mutation: 1%Indiv. replaced per generation: 50%
(Only for the 128-term equation problem)Range for new real values: (-10,10)
A single configuration
In about 24 hours, volunteer recruited from social networks visited the URL and ran the GA
256-bit Royal Road: 359 exec by 279 visitors, computing 3,952,000 evaluations128-real terms equation: 304 exec by 271 visitors, computing 4,667,000 evaluations
More than 300 executions
Browsers Operating Systems
w w
w .
g e
n e
u r
a . o
r g