Efficient algorithms for inferences on Grassmann manifolds

摘要: Linear representations and linear dimension reduction techniques are very common in signal image processing. Many such applications reduce to solving problems of stochastic optimizations or statistical inferences on the set all subspaces, i.e. a Grassmann manifold. Central them is computation an "exponential" map (for constructing geodesies) its inverse Grassmannian. Here we suggest efficient for these two steps illustrate applications: (i) For image-based object recognition, define seek optimal representation using Metropolis-Hastings type, search algorithm manifold, (ii) inferences, sample statistics, as mean variances,