This month John Langford is returning to melt faces with "Terascale Learning". Here's the abstract for the talk:
Many groups have successfully applied multiprocessor computation to deal with computationally intense learning algorithms. Unfortunately, these techniques generally do not deal with large quantities of data, which remains difficult for nontrivial learning algorithms. I will discuss new approaches for doing this, including how to learn on a 2 terafeature dataset in about an hour. These algorithms are a part of the newest release of Vowpal Wabbit ( http://github.com/JohnLangford/vowpal_wabbit/wiki ), and hence are immediately useful for either dealing with large datasets or shortening debug cycles.
John works at Yahoo! Research and has an excellent blog on machine learning at http://hunch.net/
Note about location: This month we're going to be on the 5th floor of AOL instead of the usual 6th.