C[omp]ute

From: andrew cooke <andrew@...>

Date: Sat, 20 Dec 2014 08:23:03 -0300

http://www.technologyreview.com/view/533496/why-neural-networks-look-set-to-thrash-the-best-human-go-players-for-the-first-time/

Apparently there's a database of 160,000 games between experts.  The results
are comparable to search-based approaches.  But it's not clear to me how they
improve further.  Presumably that's exhausted the available databases.  So
what next?  Some kind of hybrid approach?  If so, does the net go above or
below?  Above meaning that you have some kind of feature classification as
input; below meaning some kind of search on top.  Or something else?

Paper http://arxiv.org/abs/1412.3409

Andrew