Nov 20 2020

New Theory for Active Machine Learning in High-Dimensional and Nonparametric Hypothesis Spaces

ECE 595 Department Distinguished Lecturer Seminar Series

November 20, 2020

11:00 AM - 12:00 PM




Chicago, IL 60607

New Theory for Active Machine Learning in High-Dimensional and Nonparametric Hypothesis Spaces

Presenter: Robert A. Nowak, University of Wisconsin-Madison

Abstract: Machine learning has advanced by leaps and bounds, but mostly in well-defined domains using huge amounts of human-labeled training data. Many application domains may not be well defined and, while often data rich, they tend to be label poor.  The cost of humans labeling of large training sets can be a major bottleneck in the development of machine learning systems. Active machine learning aims to address this using algorithms to automatically select the most informative unlabeled training examples for human labeling so that their time is not wasted on irrelevant, redundant, or trivial examples.  This talk describes two fundamental challenges arising in the theory and practice of active learning: 1) designing computationally efficient strategies for selecting examples in high-dimensional feature spaces; 2) devising active learning strategies in nonparametric settings including Reproducing Kernel Hilbert Spaces and spaces of overparameterized neural networks.  Representer theorems provide a unifying approach to active learning in nonparametric hypothesis spaces, and a new representer theorem for single hidden-layer neural networks is crucial to this development.

Speaker bio: Robert Nowak is the Nosbusch Professor in Engineering at the University of Wisconsin-Madison.  He leads the Machines, Algorithms, and Data Lab (MADLab), an Air Force sponsored University Center of Excellence focused on efficient and robust machine learning.  His research and teaching interests range from machine learning and signal processing to optimization and statistics.

Faculty host: Mesrob I. Ohannessian,

Note: this seminar will not be recorded


Department of Electrical and Computer Engineering

Date posted

Nov 17, 2020

Date updated

Nov 20, 2020