n_nelson@dslextreme.com wrote:

> Attached is a zip of the demo for the Simulated Annealing program.

> 

> Anneal1.xls contains before and after graphs of the 1000 points

> used in the demo.  This is the Traveling Salesman Problem were

> the goal is to minimize the travel distance in going from point-to-

> point, each point only once, for all the points.



Have you seen Point2Point work? =P It does the same thing, with a GUI

option of entering the points.



> Simulated Annealing is a pattern reduction method in the area

> of Neural Networks, Genetic Algorithms, and statistical

> correlation and these are the basic learning programs of AI.

> 

> What we might do is assemble code for these various areas

> and then begin to apply that code to other projects such as

> NLU.  And then this is also where our distributed processing

> can be applied.  These kinds of programs may run a long time,

> use a good amount of CPU, and the methods are such that

> we can break the processing up and spread it among several

> Internet connected computers. And for this last, we might think

> about putting together a demo of Simulated Annealing on the

> Traveling Salesman problem for, say, a million points. Not

> that the Traveling Salesman problem is of any practical use in

> itself, but that we then would have the basic dstributed processing

> organization for AI learning.



Today, now that I've got my internet access back, I am going to check

out the various 'AI' chatterboxes around (like Alice, etc) and ask the

authors if it's possible to add another interface to their engines, to

change NLU-to-NLU into ICI-to-NLU and NLU-to-ICI, then see what I can do

or if I can get the very authors to cooperate (usualy they just go like

"here's the source code, do whatever you want with it").



What I see right now is that from a commercial POV, we could use

something like EWolf to make ICI more friendly to users, while from the

philosophical/academic/devolopment POV, we need a whole lot of problem

solver programs all over the place. I was thinking of integrating what

we currently have into a system, which will be perfectly aware of what

is going on around it and be able to give non-stupid responses.



The situation is such that we have a bunch of programs that could very

well do something really cool trough EWolf, only if we could get them

to:



1. Communicate a bit more (Console's EWolf support).



2. Be able to parse NLU into best-fits to any function's command set

(not all that hard as NLU might have seemed originaly, but would need

some hard brain work indeed, I urge you to look for practical

solutions).



3. Be able to conclude what is sensible and what an overkill (a nice

EXAMPLE is the Console's user message display that concludes what is

fast enough for the user to read and what is not, but I am thinking of

adding much more such features into EWolf, which will then decide when

is it reasonable to provide a message of user help, when to display a

webpage, etc).



I can assure you ICI is born for these kind of things. All of what I

speak of above has very practical solutions already, as I have been

working towards that goal all along. Now is the time to take some purely

practical action.





C'ya!



--

Cellphone: 0038640809676 (SMS enabled)



Don't feel bad about asking/telling me anything, I will always gladly

reply.

[AC/HFA(AS) -- no suprize]



Digging for info? Try AI Meta Search:

Http://WWW.AIMetaSearch.Com



MesonAI -- If nobody else wants to do it, why shouldn't we?(TM)

Http://WWW.MesonAI.Com

