9TB in 20 minutes? Thanks, IBM
theregister.co.uk
theregister.co.uk
This one gives a much better explanation, while still being accessible: http://www.hpcwire.com/features/IBM-Invents-Short-Cut-to-Ass...
Roughly speaking, when you have a lot of uncertainty on a lot of random variables it's too hard to propagate the uncertainty through the entire model. The standard approach is to just run 'enough' iterations of a monte carlo simulation - while IBM seems to have come up with a more efficient solution.
"The development of the algorithm was performed at IBM Research - Zurich and was presented on Thursday at the Society for Industrial and Applied Mathematics conference in Seattle. "
In the next, it is "software" and subject to patenting:
"The Zurich team has been working on the software for the last year-and-a-half and they were able to patent it at the end of 2009, prior to publishing the results."
So confusing... I didn't think patenting of algorithms was permitted. Color me confused...
Wasn't RSA patented, for instance?
A quick search on the author revealed the following:
GCG was founded in late 2001 by Dan Olds and is based in Beaverton, Oregon. The company focuses on enterprise information technology, including TCO (Total Cost of Ownership), IT virtualization, servers, server consolidation, and a wide variety of other issues that explore the intersections between business and technology. The goal of our research deliverables is to de-mystify technology issues and concepts so that they can be readily understood by a non-technical, business-centric audience. [1]
This doesn't look like very effective self-promotion at all, which I thought was part of the reason why people submit these articles.
[1] http://www.gabrielconsultinggroup.com/about-topmenu-39.html
That is, if you have a data table of floating point measurements with N columns and M rows, and you interpret each row as a vector in N-space, you want to know roughly how much scatter there is of the M points in N-space.
The dimensions of the covariance ellipsoid give you a good first cut at this (assuming that deviations aren't too large).
(Google results for "covariance ellipsoid" are surprisingly bad. This book result is ok though:
http://books.google.com/books?id=XphcKk7VBqUC&pg=PA116&#...)
Instead, link to the paper itself: http://portal.acm.org/citation.cfm?id=1645421&coll=GUIDE...
Which doesn't open for me. But one commenter states that they calculated an inverse covariance matrix in n^2 and made the method scale well on a parallel computer