🚀 go-pugleaf

Erik Vick wrote:

> I am interested in knowing why a neural net must be biologically
> plausible?  (Yes I'm a NN newbie)  I thought there were two accepted
> methodologies of Neural Nets : 1) the biologic modeling approach and 2)
> more of a functional modeling approach.
> 
> Is it your position that to gain common sense / learn, the construct
> must be a biologically plausible neural net?

Learning in ANNs has recently been proven to be NP-Complete (the book is
by Judd, and I can't remember the name ... someone helpful might know
it), so they ain't viable as they stand. Neuroscientists are quite
chuffed about this, because they've been telling the ANN community for
years that their models are biologically implausible.

It depends on what you want to use it for. ANNs have two possible uses: 
1) They are of practical use as pattern recognisers.
2) They produce elucidation of the way a brain functions.

Now, if learning in ANNs is NP-Complete, they certainly can't fulfill
use 1, and since its doubtful that the learning algorithm used in the
brain is NP-Complete, they pretty likely can't fulfill use 2.

Therefore, what we should like to see is ANNs based on a more complete
neurological understanding (in simple words: listen to the
neuroscientists).

I was involved in a neuroscience course in my third year at college
where they were talking about :

1) Synapses being a basic unit of computation (which the neuron was
originally thought to be), and being networked through dendritic/axonic
networks with the type of connectivity that would overcome
NP-Completeness in the learning algorithm.
2) Neurons fulfilling a temporal function, not computational

I had been involved in ANN research before going to college (I was
lucky), but now wanted a slightly different job, and so maintained an
interest in the field, but have not worked for about six years.

To be honest, I thought it was going nowhere BEFORE I read Judd's book
and took a Neuroscience course (overfitting, local minima, etc) ...
these just told me WHY.

Interestingly, Louis' method appears to be going on exactly the kind of
lines that I thought was necessary, all that time ago.

If I had never read Kuhn, I would have completely failed to understand
why this did not happen sooner.

Cheers,
walrus.