Diffusion Maps & Dimensionality Reduction

Diffusion maps build a Markov chain on point clouds via a heat kernel W_{ij} = exp(−d²_{ij}/ε). The top eigenvectors of the normalized kernel approximate Laplace-Beltrami eigenfunctions on the underlying manifold, revealing intrinsic geometry regardless of embedding.

Parameters

Eigenvalues:
λ₁ =
λ₂ =
λ₃ =
Spectral gap =