Speaker: Mikhail Belkin
Affiliation: Ohio State University
ABSTRACT: Despite remarkable recent practical successes of machine learning and large amounts of theoretical work, we are still far from fully understanding the science of data. Phenomena related to the shape of data and to the subtle but essential interactions between computation and learning are only starting to be explored. Deeper understanding of these issues is necessary for further progress of machine learning. In this talk I will discuss some of the implications of the mathematical idea of smoothness, which is arguably at the foundation of learning from data and perception. In particular, I will discuss how smoothness helps us to make use of the shape of the data, and to understand mathematical models for the data. I will also show how smoothness affects computation necessary for dealing with large data limiting what can be achieved with standard shallow methods. Understanding and addressing these limitations leads to new practical algorithms, which significantly outperform typical kernel methods while on a very modest computational budget (a few hours on a single GPU compared to supercomputer nodes or tens/hundreds of large AWS instances). Finally I will finish with some thoughts on how smoothness can explain the success of deep neural networks and whether shallow methods can achieve similar performance. BIO: Mikhail Belkin is a Professor in the departments of Computer Science and Engineering and Statistics at the Ohio State University. He received a PhD in mathematics from the University of Chicago in 2003. His research focuses on understanding the fundamental structure in data, the principles of recovering these structures and their computational, mathematical and statistical properties. This understanding, in turn, leads to algorithms for dealing with real-world data. His work includes algorithms such as Laplacian Eigenmaps and Manifold Regularization based on ideas of classical differential geometry, which have been widely used for analyzing non-linear high-dimensional data. He has done work on spectral methods, Gaussian mixture models, kernel methods and applications. Prof. Belkin is a recipient of an NSF Career Award and a number of best paper and other awards and has served on the editorial boards of the Journal of Machine Learning Research and IEEE PAMI.
Hosted by Professor Jens Palsberg
REFRESHMENTS at 3:45 pm, SPEAKER at 4:15 pm
Date(s) - Apr 06, 2017
4:15 pm - 5:45 pm