Neural Network learning (recorded Java Applet)

"The task of this neural net is to span its map over all blue points (the input values of the net) in an even way and without crossing one point twice.
This problem is similar to the classic Travelling Salesman Problem, where the shortest path between a certain number of cities is to be found without passing one city twice.
Using conventional methods, like the backtracking algorithm, this problem can't be solved on any computer in time for a number of cities greater than about 25"


To learn more about neural networks or to see the applet look yourself at: http://fbim.fh-regensburg.de/~saj39122/jfroehl/diplom/e-index.html

Category:

Education

Tags:

• neural
• network
• computer
• science
• demonstration

[zweefvlieg]

Great beat! What is it? (I like the neural part too ;-))

[mfunkmann]

Fairmont - Gazebo

[lucb1e]

ffs shut the turning i can see it from one angle as well Oo

[artilekt]

matlab is ocmercial shit :)

There are a lot of free better stuff like scillab.. or pylab

I RLY don't understand why matlab is so popular on universites?

[jeetendrag10acc2]

theres lot of free stuff,but i doubt they are better than matlab or mathematica.

I could make a complete machine control software in 20 lines of code in MAtlab.

Scilab or Pylab ,and most free alternatives for other commercial software usually have a lack of seamless integration like commercial solutions.

[TJFFx]

@artilekt Because it's a industry standard...

[marek007able]

@artilekt Plus QT Octave

[phatsmanto99]

what is the music?

[Fusionicon]

When programming a java neural net program do you use a neural net library and is so where do you download it.

[rubbercraft]

comment typo fix: "- like your target points that are relatively evenly spread"

[rubbercraft]

And I think if you had a 4 view visualization of 4 perspectives, it would reveal algorithm behavior to viewers more easily, because the random rotations are very distracting for teaching / illustrative purposes.

[rubbercraft]

Plus, the convergence of the grid apparently shows some undesirable folding of multiple gridpoints on the same target point. Better Kohonnen mappers often try to make the grid smoothly conformal, when the points are not clustered at heirarchical scales, unlike your target points that are relatively evenly spread.

[rubbercraft]

WOW, it looks like a very slow neural network algorithm for it's use of tens of hundreds of learning cycles with non-monotonic convergence. Maybe if you used NN radial basis functions and cluster convex hull characterization the convergence could be faster. It is not like it has to even worry about degenerate point clustering with multiple overlapping points of heirarchically packed densities like little solar systems spread in a loose open cluster of stars.