[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [computer-go] Automatic acquisition of syntactic structures



This was a new paper for me although I have some limited experience with
this type of papers about statistical analysis of sentences. It is a
little hard to read, since it is assumed that the reader is familiar
with the field (typical of scientific papers of course). 

The hard part is to judge if this is amazing or just another model that
looks nice but does not do anything useful. The authors show that if you
take a simple grammar and create sentences with this grammar, the model
is good enough to rediscover the grammatical structure. The model is
also shown to work for real world data but no data is provided that it
actually is good, or can compete with other methods. At least it is hard
to judge if you are not an expert in the field I guess.

Now if we apply this to go, there is an appeal to the idea that there is
a grammar of go that could be discovered by this method and apply use
this go-grammar to new positions in order to identify good moves.

One question is if this method would add something to simple brute force
pattern extraction approach. It seems to me that it is might provide
generalization in all those cases where unseen patterns are found that
the simple pattern extraction approach cannot handle at all. 

The practical problem here is to decide what the inputs to the model
should be.

Should it be sequences of moves or local patterns or something else? And
given some input what are we going to do with the output? 

Even if it does a good work in predicting which moves are good or bad in
some sense it has to be used effectively in a program that plays go,
that could not be done better or more efficiently with some other
method. 

That's my thoughts.
Magnus Persson


--
Magnus Persson
Center for Adaptive Behavior and Cognition
Tel: +49-(30)-82406-350
Cell phone: +49 163 6639868
> -----Original Message-----
> From: computer-go-bounces@xxxxxxxxxxxxxxxxx [mailto:computer-go-
> bounces@xxxxxxxxxxxxxxxxx] On Behalf Of Bob Myers
> Sent: Dienstag, 20. September 2005 01:05
> To: computer-go
> Subject: [computer-go] Automatic acquisition of syntactic structures
> 
> I wonder if any computer go researchers have seen this paper
> 
> http://www.tau.ac.il/~zsolan/papers/soletalb2002.pdf
> 
> entitled "Automatic acquisition and efficient representation of
> syntactic structures", and claiming to be an algorithm for learning
> "sequences", not only language-based but also "sheet music or protein
> sequences".
> 
> Thoughts?
> 
> --
> Bob Myers
> 
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx
> http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/