Mixture of Geodesic Factor Analyzers on Riemannian Homogeneous Spaces
arXiv:2608.06971
2026
Architecture
1 ideas extracted · analyzed Aug 31, 2026
What the math gives to ML
The paper's transferable contribution is a low-rank latent factor model whose additive reconstruction is replaced by successive Riemannian exponential maps. This gives a principled way to inject anisotropic, curved latent variation into manifold-valued representations instead of forcing them through Euclidean Gaussian layers or isotropic radial noise. A practical neural adaptation is a geodesic factor bottleneck or mixture-of-geodesic-experts layer, with learnable base points and tangent-space loading directions, tested first on spherical or SPD embeddings where exponential maps are available and numerically stable.
Ideas from this paper
Unverified
2026
Replace a Euclidean low-rank latent decoder with a geodesic factor decoder on a Riemannian manifold. A learned location α provides the component center, a small set of tangent loading vectors V captures anisotropic variation, and latent coefficients z generate curved manifold-valued features through the exponential map. Multiple such decoders can form a mixture-of-geodesic-experts layer for multimodal representations.
Useful6/10
Difficulty5/10
Novelty7/10