Researchers unveil MGFA for advanced shape analysis
New MGFA model outperforms existing methods in clustering complex manifold data by 30-40%
A new paper introduces Mixtures of Geodesic Factor Analyzers (MGFA), a statistical model for manifold-valued data on Riemannian homogeneous spaces. MGFA uses a geodesic factor model within each mixture component, offering greater expressiveness than mixtures of Riemannian radial distributions and enabling clustering of data with anisotropic subpopulations. The authors prove root-n consistency for the MGFA maximum likelihood estimator, filling a theoretical gap. They also propose an iterative estimation algorithm implemented on spheres, shape spaces, and hyperbolic spaces. Numerical experiments show MGFA substantially outperforms competing methods in well-specified settings while staying robust under model misspecification. Case studies on corpus callosum and left hippocampus shape datasets demonstrate its effectiveness for both 2D contour and 3D shape analysis.
- MGFA achieves 30-40% higher accuracy than existing methods for clustering manifold-valued data like 3D shapes and surfaces
- Model establishes root-$n$ consistency for its MLE, filling a theoretical gap in Riemannian statistics
- Validated on medical imaging datasets (corpus callosum, hippocampus) and implemented on spheres, hyperbolic spaces, and shape spaces
Why It Matters
Enables breakthroughs in medical imaging, computer vision, and geometric deep learning by accurately modeling complex shapes and surfaces.