Latent-Anything Theory

It is a sub-project to research all theory related to Latent-Anything Project.

It follows the Theory Index (THEORY.md) — the master roadmap. (THEORY.md lives at the repository root, outside this docs site, so it is linked on GitHub rather than as an internal page.)

Mục lục

• 1. Biểu diễn không gian — metric space, Mahalanobis, manifold hypothesis, curse of dimensionality, geodesic, pullback metric, FlatVI.
• 2. Representation Learning — information bottleneck, autoencoder, VAE/β-VAE, VQ-VAE, VQGAN.
• 3. Hình học & Cấu trúc — linear structure, disentanglement, isotropy/anisotropy, Riemannian geometry, slerp, normalizing flows.
• 3B. 3D Representation — neural implicit representation, NeRF, volume rendering, positional encoding, Instant-NGP, 3D Gaussian Splatting, covariance matrix, spherical harmonics, Gaussian rasterization, Gaussian-as-latent, dynamic 3DGS, Gaussian set operations.