Mean Curvature Boundary Points: A Geometric Unsupervised Learning Framework
Researchers introduce Mean Curvature Boundary Points (MCBP), a geometric framework for boundary detection in unsupervised learning. The method models intrinsic curvature of data manifolds using discrete approximation of the shape operator from local k-nearest neighbor patches. Mean curvature serves as a descriptor for boundary structure, identifying transitions between clusters and low-density interfaces. The approach departs from traditional density-based methods, offering a unified geometric interpretation of boundaries and outliers. The paper is available on arXiv under ID 2605.04274.
Key facts
- MCBP is a novel geometric framework for boundary detection
- It is grounded in Geometric Machine Learning
- The method uses discrete approximation of the shape operator
- Mean curvature is computed from local k-nearest neighbor patches
- High-curvature regions correspond to cluster transitions and irregularities
- The approach does not require explicit manifold parametrization
- It departs from traditional density-based methods
- The paper is available on arXiv with ID 2605.04274
Entities
Institutions
- arXiv