Loading Events

« All Events

  • This event has passed.

@Intuit!! Randomized algorithms for matrices and data

March 26, 2012 @ 6:30 pm

NEW LOCATION:

Intuit – MTV-09 Innovation and Invention
2600 Casey Ave
Mountain View, CA 94043

Speaker: Michael W. Mahoney

New Location Details: GETTING TO THE MEETING ROOM

Go to the sfbayACM.org web site, click on the goolge maps link, change to satellite view.    As you drive down the road, it ends in a small circle drive.  Our building is on the left.  Follow a sidewalk from the circle drive to the entrance of the left building.   From this entrance, the meeting room is close, security will guide us.  

People can park anywhere around either of the two buildings, any open spot.

Details about the talk itself:

Randomized algorithms for very large matrix problems (such as matrixmultiplication, least-squares regression, the Singular ValueDecomposition, etc.) have received a great deal of attention in recentyears. Much of this work was motivated by problems in large-scaledata analysis; this approach provides a novel paradigm andcomplementary perspective to traditional numerical linear algebraapproaches to matrix computations; and the success of this line ofwork opens the possibility of performing matrix-based computationswith truly massive data sets. Originating within theoretical computerscience, this work was subsequently extended and applied in importantways by researchers from numerical linear algebra, statistics, appliedmathematics, data analysis, and machine learning, as well as domainscientists.In this talk, we will provide an overview of this approach, with anemphasis on a few simple core ideas that underlie not only recenttheoretical advances but also the usefulness of these tools inlarge-scale data analysis applications. Crucial in this context isthe connection with the concept of statistical leverage.Historically, this notion, and in particular the diagonal elements ofthe so-called hat matrix, has been used in regression diagnostics toidentify errors and outliers. Recently, however, the connection withstatistical leverage has proved crucial in the development of improvedmatrix algorithms that come with worst-case guarantees, that areamenable to high-quality numerical implementation, and that are alsouseful to domain scientists. These developments, how to approximatevery precisely the statistical leverage scores in time qualitativelyfaster than the usual naive method, and an example of how these ideascan be applied in large-scale distributed and parallel computationalenvironments will all be described.

NEW LOCATION:

Intuit – MTV-09 Innovation and Invention

2600 Casey Ave

Mountain View, CA 94043

 

Details

Date:
March 26, 2012
Time:
6:30 pm
Event Category:
Website:
https://www.meetup.com/SF-Bay-ACM/events/56169662/

Organizer

SF Bay ACM Chapter
Website:
https://www.meetup.com/SF-Bay-ACM/

Venue

Intuit – MTV-09 Innovation and Invention
2600 Casey Ave
Mountain View, CA 94043 US
+ Google Map



Login with Linkedin