Pattern recognition Algorithm

[Arkimedes]

I have glossed over some texts on artificial neural networks, under the advice of my friend, and I have come to a problem...how does something recognize a pattern? My family has had several shrinks/neuroscientists and I recollect that there are three signs to intelligent life: the "fight-or-flight" mechanism, the ability to recognize and store patterns, and the ability to readjust the parameters of those patterns (e.g. if a stairwell had uneven stairs, the ability to readjust the parameters for that particular stairwell).

Well, the fight-or-flight mechanism is understood. On the other hand, pattern recognition is not really as well known.

So, for my own pleasure (and some help to my friend), I need to know an algorithm for recognizing patterns. I'm lazy, so rather than invent one, I'm hoping someone else has; does anyone really know anything about any pattern recognition algorithm?

[sec_ware]

Hi

Arkimedes , both your theoretical threads (this one and lambda calculus)
are interesting, but you might present your
questions in a more security related context (since it is possible). I assume you
have this in the back of your head, otherwise you won't post these questions
here on AO. Rather than Cosmos, you could have posted this question in
IDS & Scanner Discussions.

Depending on the pattern you want to recognise, there are "deterministic" ways
to do so (flow chart, formula). Let us here, however, focus on neural networks (NN).
Have a look at an older AO thread[1] with a reference to the Neural Network
Toolbox[2] for Matlab.

The common algorithmic technique is called back propagation[3] with its two variants:
forward learning and backward re-weighting. It is easy to train the NN for pattern
recognition - since you can generate a set of known patterns.
Interested in graphical pattern recognition?
first, generate black/white plots of simple geometric objects, grey/white plots
(no colour needed), enhance the complexity of the objects, but make sure that your
generator program first defines the shape, then produces the input file of the NN
I considered it simplest to overlap basic objects (circle, rectangle, triangle) to create
learn/test samples with increased difficulty.


Cheers


[1] http://www.antionline.com/showthread...hreadid=264909
[2] http://www.mathworks.com/products/neuralnet/
[3] http://ieee.uow.edu.au/~daniel/softw...h/BPN_English/
(haven't read it, but looks useful).