On Mon, Jan 20, 2014 at 3:57 AM, Ben Goertzel <[email protected]> wrote:
> > YKY, > > I would now advocate thinking in terms of the nonparametric Fisher > information... see > > *** > > Washington Mio, Dennis Badylans, and Xiuwen Liu. A com- putational > approach to fisher information geometry with appli- cations to image > analysis. EMMCVPR’05 Proceedings of the 5th international conference on > Energy Minimization Methods in Computer Vision and Pattern Recognition, > 2005. > Thanks, nice paper. I see that the non-parametric way allows to explore arbitrarily shaped probability distributions, which could be powerful. [ I now have a crude understanding of information geometry, which requires some Riemannian / differential geometry as prerequisite. ] my discussion in the attached draft (which also covers other topics)... > > To answer your question, though, in your application theta is a parameter > of a probability distribution over logic-formula space.... > 1. What is the significance of the geodesic distance? It measures the "true" distance between 2 probabilistic distributions. In your application, the search algorithm needs to advance by a distance 𝛄 (which I take is like a "learning rate" kind of parameter) in which you use the geodesic path. But how does this lead to superior performance? Even if you move in the promise landscape in a path that deviates from the geodesic, what is the disadvantage? 2. If the non-parametric info-geometry idea is applied to the space of logic formulas, that may be very interesting... ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
