Neural Networks are not Neurons
Mangala
30 Dec 2006
I'm a CS major and I was talking to my Machine Learning teacher when suddenly it dawned on me that she never mentioned neural networks a lot in class. I asked her why this was seeing as it should be the basis of learning and she told me the fundamental unit of the neural network is the perceptron, which was theorized to be a model of how the actual neuron fires. Research has shown however that neurons do not work like the perceptron in the least and are more complicated.
Can anyone point me to research on how the neuron actually works?
And given this, doesn't this show we have a much longer ways to go before we get to the SAI?
Can anyone point me to research on how the neuron actually works?
And given this, doesn't this show we have a much longer ways to go before we get to the SAI?
olaf.larsson
01 Jan 2007
I think that it is written clearly in most papers that perceptrons only have similarity to neurons and should not be regarded as a model of a neuron. The real thing has a daunting complexity I dont know how a real neuron works and I doubt anyone else knows the details for sure . If you study the subject Im very sure that you will find that there are not one kind of neuron but various subgroups of neurons that differ in their function.
Edited by wolfram, 01 January 2007 - 09:22 PM.
Edited by wolfram, 01 January 2007 - 09:22 PM.
sponsored ad
Mixter
07 Jan 2007
For one, neuron potentials trigger firings of analog spikes (spike trains between
associated neuron groups, actually) not digital/signum-function firing, so one first
step toward describing the neuron is integral-based modelling of the firing cascade.
The bad news is that's about all I really know so far, still learning quite basic stuff
to understand this sufficiently eventually:) Good news is that a good friend I know
maintains one of the most advanced, spiking neural network project which works via
analog spike trains modeled as closely to current neurology as possible (without
commercial funding, that is). Check out the project at: http://amygdala.sourceforge.net
associated neuron groups, actually) not digital/signum-function firing, so one first
step toward describing the neuron is integral-based modelling of the firing cascade.
The bad news is that's about all I really know so far, still learning quite basic stuff
to understand this sufficiently eventually:) Good news is that a good friend I know
maintains one of the most advanced, spiking neural network project which works via
analog spike trains modeled as closely to current neurology as possible (without
commercial funding, that is). Check out the project at: http://amygdala.sourceforge.net
sponsored ad
emerson
13 Jan 2007
Mixter, I'm pretty much at the base level as well. But from what I can grasp your friends work, it's awesome! And, I'm delighted by the visual module as well. I wish I wasn't so low on time right now, I'm really itching to sit down and start tinkering with this. It looks like the last cvs checkin was over a year ago, is he still working on it?
