– NESS Colloquium @ BU – Nov. 14 – David Dunson (Duke University) – Learning & Exploiting Low-dimensional Structure In High-Dimensional Data

Hi All,
The NESS Colloquium series will continue on Thursday, November 14th at Boston University with a talk Professor David Dunson from Duke University. This colloquium is open to the general public and should be accessible to those interested in statistics at all levels. The event is sponsored by the New England Statistics Society (NESS) and the Department of Mathematics and Statistics at Boston University.

Please share this email widely and please post the attached flyer. See below for more information.

Daniel Sussman
Assistant Professor, Department of Mathematics and Statistics
Boston University


Thursday, November 14, 2019
Starting at 4 pm (with refreshments served starting at 3:25 pm)


College of General Studies Building
871 Commonwealth Avenue, Boston, MA, 02215
Room CGS 129


Professor David Dunson is the Arts and Sciences Professor of Statistical Science at Duke University, has been given numerous awards including the COPSS Presidents’ Award, and is a Fellow in the Institute of Mathematical Statistics and the American Statistical Association.

Professor Dunson’s research spans numerous areas of statistics with a focus on scalable procedures with provable guarantees that can be applied to complex data structures. His work has had broad impact within the statistics community and in many other fields including biomedical research, genomics, ecology, criminal justice, and neuroscience.


Learning & Exploiting Low-dimensional Structure In High-Dimensional Data

This talk will focus on the problem of learning low-dimensional geometric structure in high-dimensional data. We allow the lower-dimensional subspace to be non-linear. There are a variety of algorithms available for “manifold learning” and non-linear dimensionality reduction, mostly relying on locally linear approximations and not providing a likelihood-based approach for inferences. We propose a new class of simple geometric dictionaries for characterizing the subspace, along with a simple optimization algorithm and a model-based approach to inference. We provide strong theory support, in terms of tight bounds on covering numbers, showing advantages of our approach relative to local linear dictionaries. These advantages are shown to carry over to practical performance in a variety of settings including manifold learning, manifold de-noising, data visualization (providing a competitor to the popular tSNE), classification (providing a competitor to deep neural networks that requires fewer training examples), and geodesic distance estimation. We additionally provide a Bayesian nonparametric methodology for inference, using a new class of kernels, which is shown to outperform current methods, such as mixtures of multivariate Gaussians.