From: lispunit_benrayfield@audivolv.com
Date: Sun Jun 13 2010 - 15:14:35 MDT
Its open-source GNU GPL 3, unless I grant multilicenses.
http://MouseSearch3d.com
I started with 1.3 million Wikipedia pages, chosen by not having any numbers in
the name and short names.
Then I used the following algorithm to reduce those to around 10000 pages:
Each iteration, count how many incoming links and outgoing links are left, and
keep only the pages that have more incoming links than outgoing links and have
at least 5 of each.
It generates a very high quality and well spread out and well connected set of
pages.
I wrote Javascript code to use it as a simulated brain. I wouldn't call it a
neural-network because it has continuous neural-activation (calculated a few
times per second), and a node only changes its own neural activation (based on
its forward and reverse links' neural activation).
Exclude reverse links that are also forward links, and forward links that are
also reverse links.
Similarly to how I narrowed down the 1.3 million pages, for each node, if the
sum of the reverse links neural activation is more than the sum of the forward
links neural activation, then decay that node's neural activation toward 3, else
decay toward -3.
A few times per second, normalize all neural activation levels on a bell curve,
average 0, standard deviation 1.
Increase neural activation when mouse touches a word on screen or gets closer or
stays close longer. Make the word flash and decay slowly.
Remove words with neural activation too close to 0, and add some more words
(with forward and reverse links) loaded from the internet (of those 10000).
Process words with very high or very low neural activation more often.
For now, the 3d part is just so you can move the words around and touch only the
words you want to touch with the mouse. It includes evolution of AffineTransform
matrixes, but the fitness function is arbitrary, just something to make it move
in interesting ways for now. Its a 6x6 matrix of x y z red green blue, and 1
offset matrix size 6.
Its simple algorithms, but it beats the other 3d search-engines that are being
created. They're building on the assumption that it should display the results
from their existing system. I built http://MouseSearch3d.com specificly for 3d
mouse interaction, as you can see in the history at
http://sourceforge.net/projects/natlangmouse
I started it as more complex code in my server, but the simple code written
above works better, so I uploaded a version that is only 1 html and
approximately 10000 text files. Why? So the people who only know Javascript can
learn AI, so it might spread to thousands of people, kind of like Mentifex's
MindForth is in an HTML file for. Almost all softwares are too complex for the
complexity of the problems they solve. Maybe if people see that small and simple
code can do complex things, they will stop buildingo n their huge ancient
infrastructure that has more problems than it solves. Most of the jobs it
appears I could get are doing the same old mindless work. That's what our
society has become. They don't know why they do it anymore. They just know it
pays the best.
Please try http://MouseSearch3d.com and tell me what you think of the
user-interface and the way it has conversations with you (with moving words
appearing and disappearing).
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:05 MDT