Distributed information-theoretic clustering

Abstract We study a novel multi-terminal source coding setup motivated by the biclustering problem. Two separate encoders observe two i.i.d. sequences $X^n$ and $Y^n$, respectively. The goal is to find rate-limited encodings $f(x^n)$ $g(z^n)$ that maximize mutual information $\textrm{I}(\,{f(X^n)};{g(Y^n)})/n$. discuss connections of this problem with hypothesis testing against independence, pattern recognition bottleneck method. Improving previous cardinality bounds for inner outer allows us thoroughly special case binary symmetric quantify gap between bound in case. Furthermore, we investigate multiple description (MD) extension CEO constraint. Surprisingly, MD-CEO permits tight single-letter characterization achievable region.