Theoretical Derivations of Min-Max Information Clustering Algorithm

The min-max information (MMI) clustering algorithm was proposed in [8] for robust detection and separation of spherical shells. In current paper, we make efforts to revisit the proposed MMI algorithm theoretically and practically. Firstly, we present the theoretical derivations of the MMI clustering algorithm, i.e., the detailed derivations of the minimization and maximization optimization of the mutual information. Secondly, several insights on the selection of the pruning parameter λ are also discussed in this paper.