The slides are pretty clear. The titles of other talks there lead to a good batch of articles to read.
The basic idea is that sampling rows and/or columns can give approximately correct matrix products very quickly. This same technique can be used to give SVD like results quickly. I see now that our internal techniques for recommendation analysis use theoretically similar (but practically quite different) techniques to make huge graph operations feasible. I guess we weren't the only ones to think along these lines. Here are some URL's: http://citeseer.ist.psu.edu/drineas01fast.html *Fast Monte-Carlo Algorithms for Approximate Matrix Multiplication* http://citeseer.ist.psu.edu/751036.html *An Experimental Evaluation of a Monte-Carlo* ... http://citeseer.ist.psu.edu/751131.html *A Randomized Singular Value Decomposition* http://www.etis.ensea.fr/~wtda/ *Workshop on Tensor Decompositions and Applications *http://citeseer.ist.psu.edu/drineas99clustering.html *Clustering in Large Graphs and Matrices* On Thu, May 22, 2008 at 6:32 PM, Allen Day <[EMAIL PROTECTED]> wrote: > My bad. There is audio for some of the talks but not any of the ones > from that week. > > -Allen > > > On Thu, May 22, 2008 at 6:13 PM, Jeff Eastman > <[EMAIL PROTECTED]> wrote: > > I found the slides, which look fascinating, but no audio for this > > presentation. Am I overlooking something? > > Jeff > > > > > > Allen Day wrote: > >> > >> There was an excellent presentation from Rob Fergus on this data set > >> at last year's UCLA/NSF "Mathematics of Search Engines" workshop. > >> > >> https://www.ipam.ucla.edu/schedule.aspx?pc=sews2 > >> > >> Scroll down to Tuesday 3pm to grab the slides and audio. > >> > >> -Allen > >> > >> > >> On Thu, May 22, 2008 at 5:44 PM, Ted Dunning <[EMAIL PROTECTED]> > >> wrote: > >> > >>> > >>> Here is a source of 1.5 million small (32 x 32) images with English > >>> labels. > >>> This might be an interesting large-ish dataset for clustering and > >>> learning > >>> experiments. > >>> > >>> The total size of the data set is pretty small (only 3+ GB) so it > >>> shouldn't > >>> be such a big deal to snag. > >>> > >>> http://people.csail.mit.edu/torralba/tinyimages/ > >>> > >>> > >>> > >>> -- > >>> ted > >>> > >>> > >> > >> > >> > >> > > > > > > > > -- > allenday.skype > +1 (415) 335-4654 (office) > +1 (310) 804-5304 (mobile) > +1 (515) 474-9337 (fax) > -- ted
