[computer-go] Re: Probable-eye heuristics

[Richard Brown]

> Please keep in mind that we are [NOT] trying for a rigorous definition of what
> an "eye" is.   

I see now.  Thanks to Don and Magnus for clarifying this point.

I hope that another few naive questions may be forgiven, to wit:

Would it be accurate to say that this approach represents
"unsupervised learning"?  That the machine-learning which
does occur within the neural networks happens, um, "on-the-fly"?

That the neural networks' predictive powers rely upon only the
observation of those random events that occur during the current
game, the current Monte Carlo simulation?

That there can be no "off-line" training of the networks?

If these things are true, then it seems to me that what is being
"learned" is more "what is the nature of the current game of go?"
as opposed to "what is the nature of expert behavior in go, in
general?"

Of course it is entirely possible that I have missed the point again.

On the flip side, I'm trying to train my neural networks using
"supervised learning" -- acquisition of go knowledge via observation
of professional behavior -- which does permit "off-line" training --
in hopes of answering the second question.

Then, having so trained the nets, I hope that a simple evaluation
of the networks will suffice to produce, for each possible behavior
in the current game, a value for the global evaluation function.

Processor speed seems far less critical under that paradigm.

What _is_ critical under that paradigm, is what we choose to measure!

I'm just trying to keep abreast of what others are doing.

If you would be so kind as to address my naive questions, I think
that I will understand more quickly, than if I educate myself in
the intricacies of Monte Carlo simulations.  I may have to do so
yet, but for the nonce I'd like to see how my own approach pans out.
"There is so much to do; so little time."

Your patience with my ignorance is appreciated.

-- 
Richard L. Brown             Office of Information Services
Senior Unix Sysadmin         University of Wisconsin System
                              780 Regent St., Rm. 246
rbrown at uwsa.edu              Madison, WI  53715