SigEvolution Newsletter

[jeff clune]

Hello all-

My paper with Hod Lipson describing the algorithm behind
EndlessForms.com is the cover article in current issue of
SigEvolution, the ACM's newsletter on Genetic and Evolutionary
Computation.

You can check it out here: http://www.sigevolution.org/

Best,
Jeff

[Amir Elaguizy]

Really enjoyed this article, it was the first SigEvolution article I'd
ever read and I've been hooked since.

I've been reading your research and am curious as to any more research
or results you have seen using the DSE you suggested in
http://jeffclune.com/publications/2011-SuchorzewskiAndClune-DSE-Gecco.pdf
?

Thanks,
Amir

[Jeff Clune]

Thanks Amir. I'm very glad you found the research interesting. 

Have you seen this video?

http://endlessforms.com/about_the_technology/

You'll probably like this paper, which summarizes a lot of my work using CPPNs to evolve neural networks for robotic control problems: 

Clune J, Stanley KO, Pennock RT, Ofria C (2011) On the performance of indirect encoding across the continuum of regularity. IEEE Transactions on Evolutionary Computation. 15(3): 346-367

http://goo.gl/qYHPR

Regarding DSE, I do not have any more results at the moment, but Marcin does have a journal paper on DSE:

Marcin Suchorzewski: Evolving scalable and modular adaptive networks with Developmental Symbolic Encoding. Evolutionary Intelligence 4(3): 145-163 (2011)

I hope that helps. Let me know if you have any further questions. Happy new year everyone!

Best regards,
Jeff Clune

Postdoctoral Fellow
Cornell University
jeff...@cornell.edu
jeffclune.com

[Amir Elaguizy]

Jeff,

Thanks! I had not seen that video although I have had some exposure to endlessforms.com as it was mentioned in your research on 3d CPPN objects.

That DSE paper was very interesting. I'm extremely interested to see if you have an exhaustive list of transfer functions, it seems like you're using standard n-ary operators (equivelance, and, or, nor, etc).

I'm interested to see if I can use the DSE or something very similar to the DSE in a spiking model using izhikevich neurons (or active dendrite) in a biologically inspired predator-prey simulation. I don't exactly know how I'd do it just yet, I'm still wrapping my head around the DSE. I'd likely have to do away entirely with transfer functions.

The concept is incredibly promising though, I think the scalability & modularity of the networks evolved by DSE offer an opportunity to study modular systems without having to manually craft the system (or at least hard wire interconnects) which removes a lot of "organic" element. This in particular is interesting to me, to see if modules form naturally around sensor inputs producing an abstraction of the sensory space which is then passed to other modules which can act upon - via unsupervised reinforcement learning. Meaning the environment and survival of the species itself is the reward mechanism.

Thanks for your research I've found it to be very inspiring and anxiously await your next paper.

Amir

[Jeff Clune]

Hello Amir,

I'm glad you like the research. Watch this space, as I am currently working on another paper regarding how to evolve modular structures in a biologically-inspired way. :-)

Your proposed project sounds both interesting and ambitious. Please keep us posted as you make progress! 

Best regards,
Jeff Clune

Postdoctoral Fellow
Cornell University
jeff...@cornell.edu
jeffclune.com