Clustering from a rate-distortion perspective
Joydeep Ghosh

Rate-distortion theory quantifies the fundamental tradeoff between
achievable compression and the fidelity of a compressed
representation. It is a general framework that allows a variety of
loss functions, and also provides insights into the model selection
problem. I shall present some recent results based on an
information-theoretic viewpoint of clustering and co-clustering using
Bregman divergences, and highlight links to prior work in this area.