[Computer-go] Parameter Optimization with Local Quadratic Regression

Jean-loup Gailly jloup at gailly.net
Thu Mar 18 18:07:44 PDT 2010


Thanks RĂ©mi for these slides, looking forward to see the paper.

> For this application it is safe to assume that the function to be
> optimized (the winning rate) is a smooth function of parameters,
> and it has no tricky local optima

Why is it safe to assume no local optimum? This is only a guess,
but with high dimension (many parameters to tune) I would expect
local optima. Does your algorithm scale to many dimensions?

Jean-loup
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20100319/4d965fb1/attachment.html>


More information about the Computer-go mailing list