IMHO, when applying artificial neural networks to an application, the structure (as well as the learning algorithm) of the network is very important. For Go, we haven't invetigated the mechanism how the brain is used yet. Backpropagation-style layered network is just a model of the cerebellum and I strongly believe we need a higher-level model to replace the modern MCTS Go programs, say, how the cerebellum works together with the other areas of the brain (such as cerebrum and basal ganglia which is said working like RL) playing a game but it's not established nor proposed. If the model approximates the mechanism of real brain well enough, it never performs well.
As a general purpose learning machine, neural networks perform much worse than sophisticated learning algorithms such as RL and also worse than suppoert vector machines, as Remi mentioned. Hideki Petr Baudis: <20091014122619.gu6...@machine.or.cz>: > Hi! > > Is there some "high-level reason" hypothesised about why there are >no successful programs using neural networks in Go? > > I'd also like to ask if someone has a research tip for some >interesting Go sub-problem that could make for a nice beginner neural >networks project. > > Thanks, -- g...@nue.ci.i.u-tokyo.ac.jp (Kato) _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
