Papers and Programming
by J.D. Marble on February 24, 2010Overview
Papers!
Efficient Allele Fitness Assignment with Self-organizing Multi-agent System [1]
Application of a Compact Genetic Algorithm to Pipe Network Optimization Problems [2]
Programming!
Progress on Extended Compact Genetic Algorithm
Problems
Results
An Idea
Allele Fitness Assignment
“The problem of discovering complex control policies for continuous tasks is often best addressed by decomposing the policy into several simpler parts.”[1]
Credit assignment problem
Multi-agent search problem
2-D pole balancing
CGA for Pipe Networks
[2]
Progress
Done:
Basic ECGA
Parametrized over genome type (bit string / real string / etc..)
To do:
Tournament selection (vs. truncation)
Partial replacement (vs. complete)
Limit building block size so 2^l \leq N (l = b.b. size, N = pop. size)
Flowchart
[3]
Problems
Enumeration of all 2^n bit combinations in a building block (out of memory)
Execution speed
Early convergence
Results So Far
Solves segmented deceptive problem (number of bits \leq 32)
(sometimes)
Idea!
Modeling the population is an optimization problem.
Try other search algorithms besides greedy, best-only
Try different heuristics for the tree search
Bibliography
[1] A. Agogino and R. Miikkulainen, “Efficient Allele Fitness Assignment with Self-organizing Multi-agent System,” Proc. Genetic and Evolutionary Computation Conference, GECCO–2004, New York, NY: Springer-Verlag, 2004.
[2] M.H. Afshar, “Application of a Compact Genetic Algorithm to Pipe Network Optimization Problems,” Civil Engineering, vol. 16, 2009, pp. 264–271.
[3] K. Sastry and D. Goldberg, “On extended compact genetic algorithm,” Genetic and Evolutionary Computation Conference, Urbana, IL, USA: 2000, p. 352–359.