Wednesday, February 10, 2016 - 4:10pm
Event Calendar Category
Other LIDS Events
Speaker Name
Trevor Campbell
Affiliation
LIDS
Building and Room number
LIDS 6th floor lounge (Bldg 32, Dreyfoos Tower)
Abstract
A curious result from extreme value theory, dubbed the Gumbel-max trick, allows us to sample a categorical random variable from an unnormalized collection of positive numbers without first normalizing them. While the result has limited application in practice (there are many better ways of actually sampling categorical random variables), it is invaluable for theoretical analysis---if the unnormalized numbers are themselves random, this trick replaces convolution with multiplication in the distribution function of the categorical random quantity. The talk will begin by discussing the importance of the result to mixture model analysis, and conclude with the introduction and proof of a new infinite extension that is applicable to modern Bayesian nonparametrics.
About LIDS Tea: LIDS Tea talks are 20 minute long informal presentations for the purpose of sharing ideas and making others aware about some of the topics that may be of interest to the LIDS audience. If you are interested in presenting in the upcoming seminars, please email Marzieh Parandehgheibi or Yasin Yazicioglu.