neural networks

2006-03-07T21:16:00.005-08:00

Neural Networks

Some links to get started:

joone
programming neural networks in java
neural networks w. java
wikipedia

what are neural networks?

-neural networks are interconnected groupings of biological neurons. In our case, we're referring to artificial neural networks, which simulate (roughly) the structure of a network of 'neurons'. This technique has been particularly useful as a means of training computers to 'learn' complex parallel processing tasks such as pattern or speech recognition, decision making and data processing.

types of neural networks:

feedforward - "In this network, the information moves in only one direction, forward, from the input nodes, through the hidden nodes (if any) and to the output nodes. There are no cycles or loops in the network." - wikipedia

Single-layer perceptron - "a single layer of output nodes; the inputs are fed directly to the outputs via a series of weights. In this way it can be considered the simplest kind of feed-forward network. The sum of the products of the weights and the inputs is calculated in each node, and if the value is above some threshold (typically 0) the neuron fires and takes the activated value (typically 1); otherwise it takes the deactivated value (typically -1). " - ibid

Recurrent networks - networks with bi-directional data flow.

there are many more types, see wikipedia for more.

basic structure:

the neuron is the core of the structure...the basic building block. It either receives input or sends output. If the neuron is above or below a particular threshold, the neuron may fire (activate) or not fire (deactivate).

here's some simple code to express this idea in java:

boolean thresholdFunction(double input)
{
if( (input>=5) && (input<=10) )
return true;
else
return false;
}

in a neural network, we must determine the number of neurons in the network (x) and the weight between this neuron and all the other neurons in the network. To do this, find the sum of all of the inputs x multiplied by the corresponding weight w.

Adjusting the individual weights alters the overall behavior of the neural network and is a very important element in the use of neural networks.

Networks work in layers...generally the top layer acting as the input layer, then a number of 'hidden' layers in between, and then finally the output layer.

Pattern Recognition:

see autoassociation

how to build a neural network example in java from Joone:

http://www.jooneworld.com/docs/sampleEngine.html

and from jeff heaton

2d graphing code

2006-02-22T12:28:00.000-08:00

here's a little code that involves taking these pixels and sending them into a PImage, as we discussed in class last week. It becomes too computationally expensive over 200 x 200. Here it's 720 by 480. The image doesn't slow down quite as much but the resolution isn't to die for. I still want to develop this code some more. I love 2d graphing. This is definitely something to explore in a LOT more depth.

http://itp.nyu.edu/~zl316/NATUREOFCODE/week5/applet/

z

2dgraphing images

2006-02-22T12:02:00.000-08:00

I'm going to try to do some large format glossy prints of these images....based on 2d graphing equations...

visual acid

2006-02-15T12:38:00.000-08:00

I love this stuff. The first one is interactive...the other is more about eyecandy:

visual acid

array of liquids

2006-02-08T12:15:00.000-08:00

here's a link to the liquid array...week 3's assignment:

http://itp.nyu.edu/~zl316/NATUREOFCODE/week3/applet/

z

gaussian distribution and arraylists

2006-01-31T15:20:00.000-08:00