General Information

A number of nonlinear dimensionality reduction, or graph embedding, techniques have been proposed recently. These embedding techniques aim to provide a low-dimensional depiction of graphs while preserving certain properties of the data. In this manuscript we propose a novel graph embedding method which tries to optimize the “modularity” of graphs during dimensionality reduction. The embedding method has a simple formulation and is naturally relaxed and solved by a convex semi-definite program, with the guaranteed global optimum. We evaluate the performance of the method with a variety of examples and the method reports promising results in inspecting the cluster structures of graphs.