A neighborhood selection algorithm for manifold learning

Manifold learning is an effective algorithm for nonlinear dimensionality reduction. The key problem of manifold learning is to confirm the neighborhood relation. For a given range, if the points can approximate a hyper plane, the neighborhood relation is clear, otherwise, the relation is fuzzy. In this paper, we studied the adjacent weights feedback of neighborhood points for neighborhood selection. It can support the manifold learning with the non-global uniformed neighborhood parameters. The efficiency of our algorithm is tested by the experimental results.