Kashtan2005sem

Latest revision as of 21:18, 24 June 2009

<bibtex> @article{kashtan2005sem,

 title=Template:Spontaneous evolution of modularity and network motifs,
 author={Kashtan, N. and Alon, U.},
 journal={Proceedings of the National Academy of Sciences},
 volume={102},
 number={39},
 pages={13773--13778},
 year={2005},
 publisher={National Acad Sciences}

} </bibtex> link

This paper was fantastic, and makes me sad. They are demonstrating the point that I am trying to. But this is four years old and I just found it. How long before I stop being four years behind and start finding papers that are relevant to me as they come out?

This isn't how they say it, but they show that you can evolve modules by selecting for them. The way they did it was by changing the environment in a way that encourages the preservation of subparts of previous solutions. This is essentially multilevel selection. My hope is to apply multilevel selection to engineering and policy. They don't see either, or the potential of selection at higher scales to increase the complexity of evolved solutions. They evolve something that has been done, they don't seem to mention, or maybe realize, that they can use the same technique to evolve solutions to problems whose dimensionality is too high for normal selection. Of course, if I'm going to show that, I have to figure out how.

They mention "the field of evolutionary design of engineered systems". I wish. But they do cite Lipson.

Both they and Lipson (cited in the paper: with Pollack and Suh 2002) put too much emphasis on the changing environment, and not enough on the higher scale selection. These authors recognize that, but still attribute much to dynamic env. Too much? must think.

"Several studies suggested that duplication of subsystems(Calabretta, R. and Nolfi 1998) of selection for stability (Variano, E.A. and McCoy, J.H. and Lipson 2004) or robustness(thompson layzell 2000) can promote modularity"

"however, computer evolution simulations under randomly chaning environemnts do not seem to be sufficient to produce modularity (citing lipson)"

They use Newman for modularity metric

"In each experiment, the difference bewenn the perfect soluutions for the two goals differ by two connections. This small difference explains how the population can rapidly adapt when the goasl is switched"

Interestingly, they found more motifs in modular graphs, and controlled for artifacts of topology. It is due to the dynamics. I never liked motifs...

Evolving a modular circuit under a fixed goal "We find that modularity decreased rapidly within a few tens of gernerations provided there is even a slight selection pressure for small sircuit size"

never use the phrase 'multi-scale'

Missing the point: "Our results help explain how modules are maintained, because of their role in a changing environment" Modules are maintained by higher scale selection, which they implemented with a modularly changing environment.

Number of goals seems to increase linearly with number of modules (look at the three module circuit that has four goals alternating s.t. it takes six goal changes to reach the first goal again.

questions

• I don't understand why the architecture was so constrained, in such a specific manner, with a different max number of inputs for each layer, and the four layers.
• I don't understand the specific definition of 'object': "...is defined by thre or more black pixels or one or two black pixels in the left column only"

quotes

"...computational evolution can currently generate designs for simple tasks, but has difficulty scaling up to higher complexity"

• I would love to see a citation for that claim. I see it everywhere, but I'm not sure how well it is understood. Is the exponential-increase-with-dimensionality-of-search-space argument the end of that story?

"This suggestion is based on the expectation that designs with higher modularity have higher adaptability and therefore higher survival rates inc hanging environemtns."

• Is adaptability a measurable or is that a theoretical argument?

"Over the course of many goal changes, modularly varying goals seem to guide the population towards a region of network space that contains fitness peaks for each of the goals in close proximity. This region seems to correspond to modular networks"

• This is a very concise view into their approach to the 2009 analysis

Citations to read:

At this point, I've read almost 'most' of the papers in the references

They used 'islands' in their search, referred to in Cantu-Paz 1995 <bibtex> @article{lipson2002omv,

 title=Template:On the origin of modular variation,
 author={Lipson, H. and Pollack, J.B. and Suh, N.P.},
 journal={Evolution},
 volume={56},
 number={8},
 pages={1549--1556},
 year={2002},
 publisher={BioOne}

} </bibtex> <bibtex> @conference{calabretta1998cse,

 title=Template:A case study of the evolution of modularity: towards a bridge between evolutionary biology, artificial life, neuro-and cognitive science,
 author={Calabretta, R. and Nolfi, S. and Parisi, D. and Wagner, G.P.},
 booktitle={Artificial Life VI, Proceedings of The Sixth International Conference on Artificial Life, MIT Press, Cambridge},
 year={1998}

} </bibtex> <bibtex> @article{variano2004nda,

 title=Template:Networks, dynamics, and modularity,
 author={Variano, E.A. and McCoy, J.H. and Lipson, H.},
 journal={Physical review letters},
 volume={92},
 number={18},
 pages={188701--188701},
 year={2004},
 publisher={APS}

} </bibtex> <bibtex> @conference{thompson2000ere,

 title=Template:Evolution of robustness in an electronics design,
 author={Thompson, A. and Layzell, P.},
 booktitle={Evolvable Systems: From Biology to Hardware: Third International Conference, ICES 2000, Edinburgh, Scotland, UK, April 17-19, 2000: Proceedings},
 pages={218},
 year={2000},
 organization={Springer}

} </bibtex>