Abstract:

We consider the ensemble clustering problem where the task is to `aggregate' multiple clustering solutions into a single consolidated clustering that maximizes the information shared among all given clustering solutions. First, we note that the notion of agreement under such circumstances could be better captured using an agreement measure based on a 2D string en coding rather than voting strategy based methods proposed in the literature. Using this generalization, we derive a nonlinear optimization model to maximize the new agreement measure. We show that the new optimization problem can be transformed into a strict 01 Semidefinite Program (SDP) which can then be relaxed to a polynomial time solvable SDP. Our experimental results indicate improvements not only in terms of the proposed agreement measure but also the existing agreement measures based on voting strategies.

This is joint work with Lopamudra Mukherjee, Jiming Peng, and Jinhui Xu.