You are here

Grad Receives Presidential Fellowship Award

CSE is pleased to congratulate Ph.D. candidate James Voss on receiving a President Fellowship Award. This award is the most prestigious award given by The Ohio State University Graduate School and recognizes the outstanding scholarly accomplishments and potential of graduate students entering the final phase of their dissertation research or terminal degree project. The Presidential Fellowship provides financial support so that each awardee may devote one year of full-time study to the completion of his or her dissertation or degree project unimpeded by other duties.

Mr. Voss is a fourth year PhD student working in the area of Machine Learning, co-advised by Mikhail Belkin and Luis Rademacher. He received B.S. degrees in Mathematics and Computer Science from Michigan State in 2011.  He is a

Machine learning is a subarea of artificial intelligence which uses the statistical analysis of data to make sense of the world.  Its relevance is highlighted by the success of corporate systems such as Apple's Siri which interacts with a cell phone user using speech recognition and the advent of self driving cars which can now be legally road tested in California, Michigan, Florida, and Nevada.  Such systems rely on machine learning techniques to find patterns within audio or video landscapes, to draw inferences, and to make decisions.  As machine learning applications become more prevalent in every day life, it becomes increasingly important that these systems both perform well and that they do not fail in unpredictable ways in the future.

James is interested in developing machine learning algorithms which are both effective in practice and have mathematical correctness guarantees.  His research is focused in unsupervised machine learning, which is the subarea of machine learning which finds patterns in raw data and organizes the data into more meaningful or interpretable forms.  Some broad examples of unsupervised learning tasks include clustering data into self-similar groups, beamforming in signal processing (for example taking multi-microphone of multiple speakers and isolating the speech signals of the individual speakers), and statistically modeling observed data.

James's research is focused on algorithms for such tasks.  More specifically, his work at Ohio State began in the area of signal separation using the Independent Component Analysis (ICA) model.  In this model, it is assumed that one observes a multi-dimensional signal which is a linear superposition of independent, underlying (unseen) source signals.  James works on algorithms for ICA.  In addition, he is interested in the mathematical structures which have been observed in the relatively vast and practical ICA literature, and the potential applicability of algorithms arising in ICA to other problems.

It turns out that the mathematical structure seen when using higher order statistics in ICA is also embedded in a number of other unsupervised learning problems including some clustering tasks, learning the parameters of a mixture distribution of spherical Gaussians, and certain tensor decomposition problems.  James's current research is in the design and analysis of unifying algorithmic frameworks for these and potentially other unsupervised learning problems.