Learning a simplicial structure using sparsity
- Author(s): Flynn, John Joseph
- Advisor(s): Yuille, Alan L
- et al.
We discuss an application of sparsity to manifold learning. We show that the activation patterns of an over-complete basis can be used to build a simplicial structure that reflects the geometry of a data source. This approach is effective when most of the variability of the data is explained by low dimensional geometrical structures. Then the simplicial structure can be used as a platform for local classification and regression.