Modelling real neurons

I would try learning about what you are dealing with. Despite the name, neural networks have nothing to do with the brain. Its a very old method for piecewise parallel linear regression. Biology doesnt know enough about how the brain works to even attempt to modelize it.

You could check out Eugene Izhikevich's papers. It is very in depth and technical, but he's got some good programs.

This paper is great for computing with neurons: spnet

Quote:

Original post by Steadtler I would try learning about what you are dealing with. Despite the name, neural networks have nothing to do with the brain. Its a very old method for piecewise parallel linear regression. Biology doesnt know enough about how the brain works to even attempt to modelize it.

I believe that the OP is looking at more biophysical NN models rather than typical ANNs. Biology doesn't know much about how the BRAIN works, but it does know a LOT about modeling neurons, and in fact biologists can do this very accurately.

I am very interested in this subject, and I am actually persuing it in my graduate studies. There is actually a lot of material on-line regarding simulating neurons at a more detailed level than what typically constitutes ANNs in CS. For instance, take a look at the NEURON software package, which includes lots of great tutorials on neuronal modeling and using it. If you're more interested in the chaotic mathematical properties, take a look at differential equation analysis programs such as XPP AUTO which is another software program used frequently in computational neuroscience. Again, with a quick search you should find an abundance of tutorials on its usage in the brain sciences as well. A word of caution: XPP AUTO is one of the very few math programs that can analyze systems of diffeqs well, but it requires a lot of patience and a good understanding of the mathematical systems to be of much use.

If you want to learn about modelling real neurons then I suggest you read up on these two things:

-Synaptic Time Dependant Plasticity
-Spiked Neural Networks

They are related, and STDP is one form of learning in SNNs.

Quote:

Original post by walkingcarcass I'm hoping to apply it to small structures of real neurons

You're going to connect it to a real brain?!

Artificial neural network models are highly highly abstract models of a real brain. Still, it's not that difficult to increase the realism a bit. Real neurons leak 'charge' and also get tired. If you want slighly more realism from your ANN model then you could add aspects like those to it.

Just how detailed are you wanting to get? Not only do neurons transmit electrical messages to other connected neurons, but they also transmit chemical messages to physically local neurons (regardless of connection). This'll make things considerably more complicated (not that it isn't complicated enough already [smile]), but is essentially for a biologically accurate model.

Quote:

Original post by joanusdmentia but they also transmit chemical messages to physically local neurons

According to a Scientific American article "neurons can release neurotransmitters far away from synapses." Also "glia broadcast signals across hardwired neurons," (take note, glia are not neurons but they still communicate) and "the unparalleled abilities of the human mind arise not from neurons but from the coherence of brain waves."

Clearly a shallow approach to similating a net of nuerons would be far off from the way an actual brain works. I would also imagine that to reasonably simulate an individual neuron, those factors must also be taken into account.

Another thing to note, in that article there was a suggestion that the electrical state of one neuron can influence the state (thru electro-magnetic induction of course) of surrounding neurons. The quote is "electrical signals transmitted through nearby axons are sometimes picked up as weak signals in adjacent axons." That effect was called "emphatic transmission."