davisidarta / topometry

What it does

We approximate the Laplace–Beltrami operator (LBO) by learning well-weighted similarity graphs and their Laplacian/diffusion operators. The eigenfunctions of these operators form an orthonormal basis—the spectral scaffold—that captures the dataset’s intrinsic geometry across scales. This view connects to Diffusion Maps, Laplacian Eigenmaps, and related kernel eigenmaps, and enables downstream tasks such as clustering and graph-layout optimization with geometry preserved.