Machine Learning and Friends Lunch

home
past talks
resources
conferences

Distributional Clustering and Graphical Models

Abstract

In this talk, I will introduce a family of graphical models in which
each node represents a clustering of a certain type of data (e.g. of
documents, words in these documents, documents' authors, documents'
titles etc). Such models are usually small but nevertheless inference
in such models is hard. I will present an efficient method for doing
inference in this type of model. The method has its roots in the
Information Bottleneck (Tishby et al. 1999) and Information-theoretic
Co-clustering (Dhillon et al. 2003). Our method (called Multi-way
Distributional Clustering) demonstrates excellent performance on large
real-world datasets (including the 20 Newsgroups and the Enron Email
dataset).