Hi all,
Thanks for the good discussion last week. We'll meet again this Friday
(July 29) at 10:30 in the EE conference room, 13th floor of Mudd
hall.
Since several of you I've talked to are working on approximation
algorithms for large problems (e.g., large scale robust PCA, large
scale clustering), I think one line of work that might be interesting
to discuss is on randomized approximations for large scale linear
algebra. There are a couple of nice semi-recent surveys in the area by
Halko, Martinson and Tropp, and by Mahoney
http://amath.colorado.edu/faculty/martinss/Pubs/2010_HMT_random_review.pdf
http://arxiv.org/abs/1104.5557
In the context of these works, I'll sneak in a little bit of
discussion on some recent (and wonderfully intuitive) work on operator
norm large deviations that could be useful for similar problems.
For the coming weeks, one suggestion was to discuss a few interesting
papers from the recent signal processing magazine special issue on
dimensionality reduction. I think this is a great idea; we'll do this
next time.
If you have any additional suggestions for topics, please let me know.
In any case, I hope to see you on Friday!
Best,
John