A Liquid State Machine Atom

Brian is too slow I think. Unless it can be run on the server and accessed. This is in fact a general principle for LSMs, they can be run on other machines continuously and interrogated in real time by my laptop that controls the robot. Thats the whole efficiency of the LSM idea in an importance sense.

For now, without that above, I'm going to try an on-laptop solution using

http://www.lsm.tugraz.at/pcsim/

which is Maass' python/C++ LSM simulator which I hope I can get working efficiently. An atom will contain a little random LSM for now.

or

http://organic.elis.ugent.be/installing_oger

or a more minimal one, which I should probably come to understand...

http://minds.jacobs-university.de/sites/default/files/uploads/mantas/code/minimalESN.py.txt

similarly...

http://aureservoir.sourceforge.net/

Guides on ESNs

http://minds.jacobs-university.de/sites/default/files/uploads/papers/PracticalESN.pdf

Alternatively just implement CTRNNs (Continuous time recurrent neural networks) in Python, Back to the good old days, long time since I've used them!!!

 e.g. from a version of NEAT 

http://code.google.com/p/neat-python/source/browse/trunk/neat/?r=352#neat%2Fctrnn

Lazy I know but I found this lovely piece of code which converts Izihkevich's Matlab script into a python script almost isomorphically with Numpy! 

https://bitbucket.org/juricap/ompc/src/tip/examples/izhikevich/izhikevich.py?at=default

This is so neat that I'll use the above I think!

OK, good news. The atom's with random Izhikevich networks are now implemented and are used as LSMs. This version can be download here...

https://github.com/ctf20/DarwinianNeurodynamics/commit/4b0650536c058ab7ba23d1c0a1627582cdf494f7

Also, I discovered some interesting patents today :)

http://www.google.com/patents/EP1901212A3?cl=en&dq=szathmary&hl=en&sa=X&ei=rUZsUYDjJ4nDPNqIgKgM&ved=0CDQQ6AEwAA

and

http://www.google.com/patents/US8103602?pg=PA1&dq=izhikevich&hl=en&sa=X&ei=10ZsUZPNOYW0PMW8gOgC&ved=0CEYQ6AEwBQ

The later was a paper subsequently.

The next few posts will describe the results of evolving atom molecules with Izhikevich based LSMs in them.