neural networks research group
areas
people
projects
demos
publications
software/data
Achieving High-Level Functionality through Evolutionary Complexification (2003)
Kenneth O. Stanley
and
Risto Miikkulainen
An appropriate but challenging goal for evolutionary computation (EC) is to evolve systems of biological complexity. However, specifying complex structures requires many genes, and searching for a solution in such a high-dimensional space can be intractable. In this paper, we propose a method for finding high-dimensional solutions incrementally, by starting with an initial population of very small genomes and gradually
complexifying
those genomes by adding new genes over generations. That way, search begins in an easily-optimized low-dimensional space and increments into increasingly high-dimensional spaces. We describe an existing method for implementing complexification, and further propose that combining complexification with an indirect genetic encoding, in which genes are reused in the specification of the phenotype, can lead to the discovery of highly complex solutions.
View:
PDF
,
PS
Citation:
In
Proceedings of the AAAI-2003 Spring Symposium on Computational Synthesis
, Stanford, CA, 2003. AAAI Press.
Bibtex:
@inproceedings{stanley:aaaiss03, title={Achieving High-Level Functionality through Evolutionary Complexification}, author={Kenneth O. Stanley and Risto Miikkulainen}, booktitle={Proceedings of the AAAI-2003 Spring Symposium on Computational Synthesis}, address={Stanford, CA}, publisher={AAAI Press}, url="http://nn.cs.utexas.edu/?stanley:aaaiss03", year={2003} }
People
Risto Miikkulainen
Faculty
risto [at] cs utexas edu
Kenneth Stanley
Postdoctoral Alumni
kstanley [at] cs ucf edu
Projects
NEAT: Evolving Increasingly Complex Neural Network Topologies
2000 - 2011
Software/Data
NEAT Matlab
The Matlab NEAT package contains Matlab source code for the NeuroEvolution of Augmenting Topologies method (see the orig...
2003
Areas of Interest
Evolutionary Computation
Neuroevolution
Reinforcement Learning