On Fri, Oct 31, 2008 at 10:26 AM, Matt Mahoney <[EMAIL PROTECTED]> wrote:
> --- On Fri, 10/31/08, Ben Goertzel <[EMAIL PROTECTED]> wrote: > > The question that worries me is: **What does it matter if AIXI __is__ > optimal, given that it uses infinitely many resources**?? > > Because it puts machine learning research on a firmer theoretical > foundation. For example, we know from experimental results that the longer > you train a neural network on a data set, the lower the training error will > get. But when you test it on a different set, there is an optimal amount of > training, after which results get worse. What AIXI does is explain this > observation. As the network is trained, it grows in algorithmic complexity. > The proper stopping point is when it is just complex enough to be consistent > with the training data, and no more. I agree that Occam heuristics are valuable in machine learning ... and we use them in OpenCog and Novamente as well ... but I don't agree that this pragmatic value is mathematically implied by the AIXI theorems Hutter proved ... ben g ------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244&id_secret=117534816-b15a34 Powered by Listbox: http://www.listbox.com
