Distributed Information-Theoretic Biclustering

We study a novel multi-terminal source coding setup motivated by the biclustering problem. Two separate encoders observe two stationary, memoryless sources $X^n$ and $Z^n$, respectively. The goal is to find rate-limited encodings $f(x^n)$ and $g(z^n)$ that maximize the mutual information $I(f(X^n);g(Z^n))/n$. We present non-trivial outer and inner bounds on the achievable region for this problem. These bounds are also generalized to an arbitrary collection of stationary, memoryless sources. The considered problem is intimately connected to distributed hypothesis testing against independence under communication constraints, and hence our results are expected to apply to that setting as well.