Nonlinear dimensionality reduction by semidefinite programming
Lawrence Saul
University of Pennsylvania
How can we detect low dimensional structure in high dimensional data?
Combining ideas from spectral graph theory, convex optimization,
and differential geometry, I will describe two recently
developed methods for analyzing high dimensional data that has been sampled
from a low dimensional submanifold. These methods can be
used to estimate the dimensionality of the submanifold
and to derive faithful low dimensional representations of
high dimensional data.