Manifold Learning with Geodesic Minimal Spanning Trees

Jose Costa,Alfred Hero

DOI: https://doi.org/10.48550/arXiv.cs/0307038

2003-07-17

Abstract:In the manifold learning problem one seeks to discover a smooth low dimensional surface, i.e., a manifold embedded in a higher dimensional linear vector space, based on a set of measured sample points on the surface. In this paper we consider the closely related problem of estimating the manifold's intrinsic dimension and the intrinsic entropy of the sample points. Specifically, we view the sample points as realizations of an unknown multivariate density supported on an unknown smooth manifold. We present a novel geometrical probability approach, called the geodesic-minimal-spanning-tree (GMST), to obtaining asymptotically consistent estimates of the manifold dimension and the Rényi $\alpha$-entropy of the sample density on the manifold. The GMST approach is striking in its simplicity and does not require reconstructing the manifold or estimating the multivariate density of the samples. The GMST method simply constructs a minimal spanning tree (MST) sequence using a geodesic edge matrix and uses the overall lengths of the MSTs to simultaneously estimate manifold dimension and entropy. We illustrate the GMST approach for dimension and entropy estimation of a human face dataset.

Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The problems that this paper attempts to solve mainly focus on the new physical properties and effects of wave propagation in biaxial anisotropic left - handed materials. Specifically:

1. **Left - right coupling of circularly polarized light**: In biaxial optically active left - handed materials, since both the dielectric constant and the permeability tensor of the material are negative, it leads to the coupling phenomenon between left - handed and right - handed circularly polarized light. This coupling is the result of the combined action of the optical activity and left - handed characteristics of the material.

2. **Independence of geometric phase**: Normally, the geometric phase of a photon in a bent fiber depends on the cone angle of the solid angle enclosed by the trajectory of the light wave vector direction. However, in some biaxial anisotropic left - handed media, the geometric phase of the photon is independent of the cone angle, which makes it possible to study new geometric phases.

3. **Extra phase at the interface**: When a light wave propagates from a left - handed medium to a right - handed medium or vice versa, an extra phase change will occur at the interface. These phase changes are caused by the instantaneous chirality reversal and have been studied in detail through the Lewis - Riesenfeld invariant theory.

4. **Detection of the quantum vacuum geometric phase**: By using some biaxial anisotropic left - handed media, experiments can be designed to detect the geometric phase in the quantum vacuum, which is of great significance for understanding the vacuum effects in quantum field theory.

5. **Frequency control and application**: The frequency change of circularly polarized light is studied in some inhomogeneous media (such as photonic crystals) through spatial structure control, so as to achieve a controllable position - dependent frequency shift.

In summary, this paper aims to deeply explore the unique properties of wave propagation in biaxial anisotropic left - handed materials, especially the coupling of circularly polarized light, the independence of geometric phase and the extra phase change at the interface, and proposes some potential application prospects.