6 Conclusions
In Phase One, several fitness
functions were used to evolve images demonstrating suggestive
principles. In nearly all cases, successful evolution was achieved
quickly, generating images recovering some of the complexity of
two-dimensional CAs.
In Phase Two, the Bluenome
model was applied to a non-trivial artificial problem, one which
involved the coordination of many non-linearly interacting components.
In cases of low phenotypic complexity, the bijective methodology tended
to outperform the Bluenome method, with a wide margin in initial
generations, barely so in later generations. In cases of high phenotypic
complexity, however, one of the Bluenome runs clearly outperformed all
of the bijective runs, with a second matching with potential for further
growth in later generations. The Bluenome methodology continued to
develop in a high-dimensional space, while the bijective methodology
stagnated early on.
In addition to this
performance increase in high complexity runs, the Bluenome model showed
an inherent ability to generate agents with better developed systems for
the distribution of food throughout the body - this is no doubt a result
of the inheritance of cell specialization creating a network of
transport cells; Perhaps it is an easily identified example of one of
many inherent structural properties created by the developmental
process. It is unknown what other such mechanisms might exist.
Finally, the resistance of the developmental process to changes in the
environment was demonstrated - this is an intriguing notion, that an
agent's development should respond to the developing environment and
adapt. More intriguing still is the continuation of performance by the
re-developed agents, both in terms of valuation by the fitness function
in question, and in visual appearance.
7 Future Directions
The
artificial problem used in Phase Two was a difficult problem; That, and
other similar problems might be sufficient to convince practitioners of
the utility of this system. However, to evaluate claims regarding the
utility of developmental models in general requires a more in-depth
study. A study of the inherent structure created by developmental
processes in general is in order; This is our current interest,
initially by attempting to objectively measure the notion of continuity
between genotype and phenotype.
A second matter touched
upon in the above discussions is the view of evolution as a mechanism
for controlling complex processes; Indeed, this is an intriguing
hypothesis, perhaps contributing to the success of the above system; If
true, however, it begs an obvious question: by what mechanism? It is the
hope of the authors that systems like Bluenome may serve as a test bed
by which this claim may be evaluated and studied further.
References
1. Bentley, P., Kumar, S., The Ways to
Grow Designs: A Comparison of Embryogenies for an Evolutionary Design Problem,
in Proceedings of the Genetic and Evolutionary Computation Conference,
GECCO-1999
2.
Dellaert, F., Beer, R., A Developmental Model for the Evolution of
Complete Autonomous Agents, in From Animals to Animats 4: Proceedings of
the Fourth International Conference on Simulation of Adaptive Behavior,
(1996)
3.
Eggenberger, P., Evolving Morphologies of Simulated 3D Organisms Based
on Differential Gene Expression., in Husbands, P., Harvey, I. (editors)
Proceedings of the Fourth European Conference on Artificial Life (1997)
4.
Fagotto, F., Gumbiner, B., Cell contact-dependent signaling, in
Developmental Biology 180 (1996)
5.
Furusawa, C., Kaneko, K., Emergence of Rules in Cell Society:
Differentiation, Hierarchy, and Stability, in the Bulletin of
Mathematical Biology (1998)
6.
Hamahashi, S., Kitano, H. Simulation of Drosophilae Embryogenesis, in
the Proceedings of the Sixth International Conference on Artificial Life
(1998)
7.
Hotz, P., Gomez, G., Pfiefer, R., Evolving the morphology of a neural
network for controlling a foveating retina - and its test on a real
robot, in Artificial Life VIII: The 8th International Conference on the
Simulation and Synthesis of Living Systems (2003)
8.
Karp, G., Cell and Molecular Biology, 3
rd Ed., (John Wiley &
Sons, Inc.; 2001)
9.
Kauffman, S., The Origins of Order, (Oxford University Press; 1993)
10. XXXXXXX, X., Bluenome: A Novel
Developmental Model for the Evolution of Artificial Agents, Master's Thesis,
XXXXXXXX University (2003) (I have posted a version with front matter removed
on a friend's website: http://www.karafraser.com/bluenome)
11.
Kvasnicka, V., Pospicjal, J., Emergence of Modularity in
Genotype-Phenotype Mappings, in Artificial Life, v. 8, no. 4 (2002)
12.
Mitchell, M., Crutchfield, J., Das, R., Evolving Cellular Automata with
Genetic Algorithms: A Review of Recent Work, in Proceedings of the First
International Conference on Evolutionary Computation and Its
Applications (1996)
13.
Olivera, G., Olivera, P. Omar, N. Definition and Application of a Five
Parameter Characterization of One-Dimensional Cellular Automata Rule
Space, in Artificial Life, Volume 7 Number 3, 2001
14. Stanley, K., Miikkulainen, R., A
Taxonomy for Artificial Embryogeny, in Artificial Life, v.9, no. 2 (2003)
15. Wolfram, S.: A New Kind of Science.
Wolfram Media New York (2002)
16. Wolfram, S., Universality and
Complexity in Cellular Automata. In Wolfram, S. Cellular Automata and
Complexity: Collected Papers. Westview Press (1994)
Page 1,
Page 2, Page 3
|