Prof. Jeffrey BILMES from University of Washington, Seattle, USA, will present his work

Generalized Combinatorial Independence: Submodular Summarization of Large Data Sets

on Monday, 22.09.2014, at 14:00, in our seminar room IDEG134, Inffeldgasse 16c/EG.

Abstract:

The explosion of available data is both a blessing and a curse for the field of machine learning. While bigger data is different data, and can lead to improved statistical accuracy, bigger data may also be plagued with redundancy. In this talk, we will see how submodular functions are natural for generalized combinatorial (conditional) independence functions, and how many submodular programs can be seen as data summarization strategies. We will review how submodular functions can practically be used for big data summarization, and will do so in four different domains. Starting with our earlier work on document summarization, we will next review recent work on summarization large speech corpus training data, summarizing training data for the sake of a given test data set in statistical machine translation, and lastly some work on image collection summarization.