A nonlinear filtering method for geometric subspace tracking
Anuj Srivastava · 2002
We formulate the problem of tracking principal subspaces as a problem in nonlinear filtering. The subspaces are represented by their complex projection-matrices, and moving subspaces correspond to trajectories on the Grassmann manifold. Taking a Bayesian approach, we impose a smoothness prior on the subspace rotation. Combining ideas from importance sampling and sequential methods, we apply a recursive Monte Carlo approach to solving for MMSE estimates.