Well, the other day I had an idea to make something at work work better.

It involved neural nets (lol, I like to go for the easy things).

Anyway, I started today by teaching myself genetic algorithms. It's a fairly trivial example as all it does is find how to achieve a target value by using single digit numbers and the standard operators.

Got that working this morning.

Now I'm working on the next part. I had the idea to do a talk on it at work so I came up with an idea that would work well with it.

I'm creating a neural network (within an iPhone app) that I'm going to use genetic algorithms to "teach" it how to recognise faces Well, to begin with it's only hand drawn "stick-man" faces on a 64x64 "screen".

It's actually not as hard as I always used to think it was.

I might write it up on my blog once I've done it. Essentially you take the "value" (1 or 0) from each pixel of the image and input it into each neuron in the first layer of the network. Each neuron then multiplies each input by a weight and adds them all together if this is higher than the threshold value then you output a 1 otherwise a 0. You take these outputs and put them through another layer using the same maths. Then you pass it in to the required number of output neurons and the output of that is your answer. (i.e. 1 = face, 0 = not a face).

Anyway, I'm currently building the network. The hardest part is setting up the reusable stuff for the first time.

Looking forward to getting it finished though

Have you seen this
http://news.dice.com/2012/08/23/generat ... thmically/

And a live demo here
http://iobound.com/pareidoloop/

You get results like this

That's very cool. Had a look at a few of them. The results vary a lot.

However... BOOM!

Neural net is working!

I've not got any training going through it at the moment it's just a single neural net that analyses a random array of 4096 "pixels".

So there's 2^24 interconnections between layer? That would take a while to do on a wire-wrap prototyping board if you were doing it the old way with comparators and resistors

Be interesting to compare the "intelligence" to biological examples. That's the same number of "brain cells" as a slug, but the "synapse" count is much higher up the scale - nearer that of a fly.

http://en.wikipedia.org/wiki/List_of_an ... of_neurons

https://www.coursera.org/course/neuralnets

Might be of some interest

There are more on http://www.openculture.com/free_certificate_courses but that seemed the most suitable)