Riemannian Archetypal Analysis: Interpretable non-linear data analysis on deformed star distributions

In three linesRiemannian archetypal analysis using data-driven pullback geometry on deformed star distributions. Combines interpretability of classical archetypal analysis with non-linear model expressiveness. Riemannian archetypal mapping (RAM) projects onto manifolds of geodesically convex archetype combinations. Experiments on MNIST demonstrate meaningful geodesics and geometry-aware denoising.

Summary generated by Claude — human-verified