[Ml-stat-talks] Fwd: matt taddy: monday november 11 at 1:30pm

David Blei blei at CS.Princeton.EDU
Mon Nov 11 08:07:28 EST 2013


reminder: matt taddy is speaking today at 1:30pm in CS room 301.


---------- Forwarded message ----------
From: David Blei <blei at cs.princeton.edu>
Date: Tue, Nov 5, 2013 at 9:55 PM
Subject: matt taddy: monday november 11 at 1:30pm
To: ml-stat-talks <ml-stat-talks at lists.cs.princeton.edu>


mark your calendars for next monday november 11 at 1:30pm.  matt taddy
from the university of chicago will discuss inverse regression,
including his latest JASA discussion paper "multinomial inverse
regression for text analysis".  matt has deep intuitions, interesting
results, and a fresh perspective on analyzing massive discrete data

the details are below my signature.



Inverse Regression for Analysis of Big Count Data
Matthew Taddy
University of Chicago

Monday, November 11
Computer Science, Room 301

Much of the data we encounter occur as counts in a massive number of
categories: think of word usage in text mining, or domain visits in
web analytics. This data is accompanied by a smaller set of attributes
-- associated with, say, an author or IP address -- and we want to be
able to build a model that will be useful when some of these
attributes are unobserved.  This talk will discuss multinomial inverse
regression as a strategy for projecting from high-dimensional counts
onto such a set of attributes.  We'll consider the computational and
statistical reasons one would want to do this, and consider a series
of examples on how the techniques can work (or not work) in both
inference and prediction.

More information about the Ml-stat-talks mailing list