| From: | Geoff Skellams geoff.skellams@*********.com.au |
|---|---|
| Subject: | Headware Memory |
| Date: | Thu, 11 Feb 1999 10:30:55 +1100 |
wrote:
> But who is to say that the way that a human brain works when you get
right
> down to it isn't actually binary. Neuron's firing. Effectively an
ON OFF
> type of thing. And that my friend would be Binary would it not?
On/Off or
> 1/0
That's a very simplistic view of the way a neuron works and is
not necessarily right. There is a whole branch of computer science right
now dealing with neural networks. They aim to create software (and
sometimes hardware) that mimics the way a human brain works. Only a very
simple one uses a binary threshold function (either it fires or it
doesn't, based on the inputs). The more advanced ones use some sort of
non-linear threshold functions and quite often use fuzzy logic.
For the uninitiated, fuzzy logic not only includes the concept
that you can have diametrically opposed answers (for example, true and
false), but a value can be a mixture of the two values (37% true and 63%
false). You then build up a series of rules about what you do across the
entire range of possible values. When you test something, you fire ALL
of the rules each time and calculate the average of the results. This
then gives you your answer. By building up a decent rule base, you can
do some amazing things. I read in one book that there was a Japanese
researcher who has developed a rule base of about 100 rules that can
safely land a model helicopter that loses a rotor blade (although I have
wondered if they mean loses the tail rotor). Unfortunately, I haven't
heard any more about this experiment, so I have no more details on it.
If you have neurons that work along a sliding scale like this,
then you are going to have neurons effectively half firing. Any given
neuron might take inputs from 3 different inputs. If input A is firing
to 30% capacity, input B is firing to 24% of its capacity and input C is
firing to 98%, then the target neuron might (based on the evaluation
function that it has) might decide to fire to 73% of its capacity. Your
threshold function is going to accept any combinations of its inputs and
it will fire to a certain degree that won't be necessarily the maximum
that it can do.
And that ain't binary. You can't get it into one of two classes.
I'm not even going to mention Quantum computing (mainly because
I haven't heard a lot about it, although it does sound rather scary).
cheers
G
--
Geoff Skellams R&D - Tower Software
Email Address: geoff.skellams@*********.com.au
Homepage: http://www.towersoft.com.au/staff/geoff/
ICQ Number: 2815165 (Eynowd)
Hili hewa ka mana'o ke 'ole ke kukakuka
(Ideas run wild without discussion)
