Graduate/faculty Seminar
Friday, September 18, 2009, 4:30pm, 119 Physics
Anna Little
Intrinsic Dimensionality Estimation for Data Sets
Abstract:
We consider a novel approach for estimating the intrinsic dimensionality of high-dimensional point clouds. Assuming that the points are sampled from a k-dimensional manifold embedded in R^D and corrupted by D-dimensional noise, with k << D, we estimate dimensionality via a new multiscale algorithm that generalizes PCA. The algorithm exploits the low-dimensional structure of the data, so that its power depends on k rather than D. Connections with random matrix theory will also be explored.

Generated at 7:30am Friday, March 29, 2024 by Mcal.   Top * Reload * Login