Stephen Tyree

Greetings from St. Louis! I am a Ph.D. student at Washington University in St. Louis in the Department of Computer Science and Engineering. Advised by Kilian Q. Weinberger and Kunal Agrawal, my research centers on parallel algorithms for machine learning.

My work spans a wide variety of machine learning approaches (e.g. tree methods, support vector machines, metric learning) and parallel systems (multi-cores, distributed systems, and graphics processors). Among the results of my research is WU-SVM, currently the fastest support vector machine training software for GPUs and multi-cores. I am a recipient of the 2013-2014 NVIDIA Graduate Fellowship.

I hold a Bachelors degree in computer science and mathematics and a Masters degree in computer science, both from the University of Tulsa. My previous research has included computer network threat modeling, formal computational models for cellular biology, and automated sulcus classification in anatomical brain scans. Born in Memphis, Tennessee and raised in Oklahoma, I have a natural affinity for good barbecue and the occasional thunderstorm. In St. Louis, I enjoy New City Fellowship, baseball, cooking, and playing an occasional game of soccer.

News

November 2014  Completed Dissertation Defense

I have successfully defended my doctoral dissertation, entitled "Approximation and Relaxation Approaches for Parallel and Distributed Machine Learning." I will graduate from Washington University in December 2014 and am currently on the job market. I am seeking a research and development position in machine learning or data science leveraging expertise in computer science for large-scale data analytics.

April 2014  "Stochastic Neighbor Compression" at ICML 2014 ICML 2014 logo

"Stochastic Neighbor Compression" by Matthew J. Kusner, Stephen Tyree, Kilian Q. Weinberger, and Kunal Agrawal will appear at this year's International Conference on Machine Learning (ICML) in Beijing, China. This paper introduces a novel dataset compression technique for k-nearest neighbor classification. The algorithm learns a much smaller synthetic dataset (of user-specified size) which imitates the larger training set while minimizing the 1-nearest neighbor classification error. Results demonstrate no loss in 1-NN accuracy at up to 1-2% compression across a range of datasets. This technique may be particularly advantageous in porting nearest-neighbor classification to limited-memory settings (such as mobile devices or GPUs).

March 2014  Presentation at NVIDIA GTC 2014 NVIDIA GTC logo

I presented a talk entitled "Machine Learning with GPUs: Fast Support Vector Machines without the Coding Headaches" at the NVIDIA GPU Technology Conference in San Jose, CA on March 25, 2014. We debuted WU-SVM, an extremely fast implementation of Support Vector Machine training for GPUs and multi-cores.

March 2013  Awarded NVIDIA Graduate Fellowship NVIDIA logo

I am extremely pleased and grateful to announce that I have been named one of eleven recipients of the 2013 NVIDIA Graduate Fellowship. The fellowship funds Ph.D. students "who are researching topics that will lead to major advances in the graphics and high-performance computing industries, and are investigating innovative ways of leveraging the power of the GPU." I will be supported by this award for the 2013-2014 academic year and am very excited to continue my work to scale up machine learning methods using GPU hardware.

Publications

Stephen Tyree. Approximation and Relaxation Approaches for Parallel and Distributed Machine Learning. Ph.D. Dissertation, Washington University in St. Louis, 2014.

[Paper] [Bibtex] [mlpapers] Matthew J. Kusner, Stephen Tyree, Kilian Q. Weinberger, and Kunal Agrawal. Stochastic Neighbor Compression. Proceedings of 31st International Conference on Machine Learning (ICML), pages 622-630, 2014.

Stephen Tyree, Jacob R. Gardner, Kilian Q. Weinberger, Laurens van der Maaten, and Kunal Agrawal. Random Forest Ensemble Metrics. Technical Report, 2014.

[Paper] [Code] [Bibtex] Stephen Tyree, Jacob R. Gardner, Kilian Q. Weinberger, Kunal Agrawal, and John Tran. Parallel Support Vector Machines in Practice. Technical Report, arXiv:1404.1066, 2014.

Laurens van der Maaten, Minmin Chen, Stephen Tyree, and Kilian Q. Weinberger. Learning with Marginalized Corrupted Features. In revision, Journal of Machine Learning Research, 2013.

[Paper] [Code] [Bibtex] [mlpapers] Laurens van der Maaten, Minmin Chen, Stephen Tyree, and Kilian Q. Weinberger. Learning with Marginalized Corrupted Features. Proceedings of 30th International Conference on Machine Learning (ICML), pages 410-418, 2013.

[Paper] [Code (in LMNN 2.4)] [Bibtex] [mlpapers] Dor Kedem, Stephen Tyree, Kilian Q. Weinberger, Fei Sha, and Gert Lanckriet. Non-Linear Metric Learning. In Proceedings of Advances in Neural Information Processing Systems (NIPS), pages 2582-2590, 2012.

[Code] Stephen Tyree, Kilian Q. Weinberger, Kunal Agrawal. pGBRT: Parallel Gradient Boosted Regression Trees. Technical Report, 2011.

[Paper] [Code] [Bibtex] Stephen Tyree, Kilian Q. Weinberger, Kunal Agrawal, and Jennifer Paykin. Parallel Boosted Regression Trees for Web Search Ranking. Proceedings of the 20th international conference on World Wide Web (WWW), pages 387-396, ACM, 2011.

Code

WU-SVM  Support Vector Machine Training for Multi-Cores and GPUs

[Site] [Download] [Paper] [Bibtex] This software package learns support vector machine classifiers using the Sparse Primal formulation proposed in (Keerthi, et al., 2006). Both GPU and multi-core platforms are supported. The software is LibSVM compatible, using the LibSVM command line interface and input/output formats. For more information, see the project website.

pGBRT  Parallel Gradient Boosted Regression Trees

[Site] [Download] [MLOSS] [Paper] [Bibtex] This software package learns gradient boosted regression tree ensembles. Tree learning is parallelized for multi-core and distributed systems using MPI. For more information, see the project website or the MLOSS repository.

CV

My resume is available in [PDF] and [HTML] formats.

Contact

Email
swtyree@wustl.edu

Office
Jolley Hall, Room 422

Mailing Address
Department of Computer Science and Engineering
One Brookings Drive
Campus Box 1045
St. Louis, MO 63130