The results that they are reporting don't seem state of the art at all. E.g., they report that the top-ranked move from their neural net has a 10% chance of matching a pro move on a 19x19 board.
My impression is that you can get better prediction accuracy just by using the distance to the last opponent's move. And if you use a large-scale pattern approach (e.g., Remi's work), then you can predict over 40% and maybe up to 50% of pro moves. Am I remembering this correctly? -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Mark Boon Sent: Wednesday, January 23, 2013 2:45 PM To: [email protected] Subject: Re: [Computer-go] Recursive Neural Networks Sorry, never mind. Next time I'll Google the title-author combination before bothering anyone else, rather than after. Mark On Wed, Jan 23, 2013 at 9:36 AM, Mark Boon <[email protected]> wrote: > OK, this is a bit of a cheapskate question. > > For a project I'm currently working on I'm considering looking into > recursive neural networks. I bumped into the following article > describing a project that used them to make a Go playing program: > http://www.ncbi.nlm.nih.gov/pubmed/18420381 > > I figure that even though I'm not going to use it for Go, being > familiar with computer Go may make it easier for me to understand. But > it costs $31.50 to download the article. I'm balking at the cost > because a) I think these type of articles should actually be free to > access and b) I've always been very skeptical this type of approach > would work for Go so it may actually be a completely useless article > at that. > > So I was wondering if anyone was aware of the article and could tell > me whether it's worth even a buck, let alone thirty-one of them, to > download it. > > Mark _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
