KRISTJAN GREENEWALD
RESEARCH INTERESTS Machine learning, statistical learning, signal processing, reinforcement learning. Time-varying models and algorithms for learning and exploiting structure in high dimensional data.
EDUCATION University of Michigan September 2012-January 2017 Ph.D. in Electrical Engineering (Signal Processing) Advisors: Profs. Alfred Hero III and Shuheng Zhou (Statistics Dept.)
Wright State University M.S. in Electrical Engineering (Signal Processing) September 2011-August 2012 B.S. in Electrical Engineering, Magna cum Laude September 2010-June 2011
POSITIONS IBM Research, MIT-IBM Watson AI Lab January 2018 - present Research Scientist Cambridge, MA · Research collaborating with MIT under the MIT-IBM Watson AI Lab.
Harvard University, Statistics Department September 2017 - January 2018 Postdoctoral Research Fellow Cambridge, MA · Research under Prof. Susan Murphy in the Statistical Reinforcement Learning Lab, developing bandit algorithms robust to a wide variety of real-world nonidealities typical of mobile health applications, with associated performance prediction results. · Applied to real-world personalized decision making problems for adaptive, timely behavioral interven- tions in a mobile health app.
University of Michigan, Statistics Department January 2017 - August 2017 Postdoctoral Research Fellow Ann Arbor, MI · Working with Profs. Susan Murphy and Ambuj Tewari in the Statistical Reinforcement Learning Group. · Research on developing bandit algorithms robust to a wide variety of real-world nonidealities typical of mobile health applications, with associated performance prediction results.
University of Michigan, EECS Department September 2012 - January 2017 Graduate Research Assistant Ann Arbor, MI · Dissertation Advisors: Profs. Alfred Hero III and Shuheng Zhou (Statistics). · PhD research on online dynamic learning and time-varying high dimensional covariance estimation, with associated learning algorithms and performance prediction results. · Applied to solving problems including modeling of spatio-temporal processes, human brain modeling, video/sensor network anomaly detection, and activity based clustering/segmentation of crowds.
Air Force Research Laboratory (Sensors Directorate) July - September 2015 Summer Intern Dayton, OH · Developed GPU implementation of low-rank Kronecker PCA algorithms for covariance based spatio- temporal clutter removal and target detection in synthetic aperture radar imagery.
MIT Lincoln Laboratory May - July 2015 Summer Intern Lexington, MA · Developed adaptive online learning algorithms with strong performance guarantees for learning optimal data metrics in dynamic environments, with applications to semisupervised clustering and classification.
Air Force Research Laboratory (Sensors Directorate) May - August 2014 Summer Intern Dayton, OH · Developed low-rank Kronecker PCA algorithms for covariance based spatio-temporal clutter removal and target detection in synthetic aperture radar imagery.
Air Force Research Laboratory (Sensors Directorate) June - August 2013 Summer Intern Dayton, OH · Developed algorithms for the detection of crowd behavior anomalies in video, based on efficient learning of sparse and low rank high-dimensional spatio-temporal models.
Wright State University, EE Department September 2011 - August 2012 Graduate Research Assistant Dayton, OH · Thesis Advisor: Prof. Brian Rigling. · MS thesis research in the area of Laser Radar target classification performance prediction. · Worked as intern at AFRL Sensors Directorate June - August 2012.
Air Force Research Laboratory (Sensors Directorate) June 2011 - August 2011 Summer Intern Dayton, OH · Worked with developing/improving software to efficiently create physically accurate, model-based syn- thetic Flash Laser Radar imagery. · Part of a team that developed a unified multi-sensor synthetic imagery generation system to enable more efficient evaluation of exploitation algorithms.
Wright State University, EE Department September 2010 - June 2011 Undergraduate Research Assistant Dayton, OH · Research Advisor: Prof. Brian Rigling. · Developed algorithms to compute 3D imagery from sets of 2D images using tomographic backprojection followed by expectation maximization.
Air Force Research Laboratory (Sensors Directorate) June 2010 - August 2010 Summer Intern Dayton, OH · Worked on radar image formation with work involving implementation of the monostatic Polar Format Algorithm for Synthetic Aperture Radar (SAR) as a MATLAB program, as well as working with several other SAR image formation algorithms.
PUBLICATIONS Z. Goldfeld, E. Van den Berg, K. Greenewald, B. Kingsbury, I. Melnyk, N. Nguyen, and Y. Polyanskiy, “Estimating Information Flow in DNNs,” submitted to ICML, 2019. M. Yurochkin, M. Agarwal, S. Ghosh, K. Greenewald, N. Hoang, and Y. Khazaeni, “Bayesian Non- parametric Federated Learning of Neural Networks,” submitted to ICML, 2019. Z. Goldfeld, K. Greenewald, J. Weed, and Y. Polyanskiy, “Estimating Differential Entropy under Gaus- sian Convolutions,” submitted to IEEE Transactions on Information Theory, 2018. P. Liao, K. Greenewald, P. Klasnja, and S. Murphy, “Personalized HeartSteps: A Reinforcement Learn- ing Algorithm for Optimizing Physical Activity,” submitted to IMWUT, 2018. K. Greenewald, S. Zhou, and A. Hero, “Tensor Graphical Lasso (TeraLasso),” Under revision at Journal of the Royal Statistical Society, Series B, available as arXiv, 2017. Z. Goldfeld, K. Greenewald, J. Weed, and Y. Polyanskiy, “Optimality of the Plug-in Estimator for Differential Entropy Estimation under Gaussian Convolutions,” IEEE International Symposium on Information Theory, 2019. K. Moon, K. Sricharan, K. Greenewald, and A. Hero, “Ensemble Estimation of Information Divergence,” Entropy, 2018. K. Greenewald, A. Tewari, P. Klasnja, and S. Murphy, “Action Centered Contextual Bandits,” NIPS, 2017. K. Greenewald, S. Park, A. Giessing, and S. Zhou, “Time-dependent spatially varying graphical models, with application to brain fMRI data analysis,” NIPS, 2017. K. Greenewald, S. Kelley, B. Oselio, and A. Hero, “Similarity Function Tracking Using Pairwise Com- parisons,” IEEE Transactions on Signal Processing, 2017. K. Greenewald, “High Dimensional Covariance Estimation for Spatio-Temporal Processes,” PhD Thesis, January 2017. K. Greenewald, S. Kelley, and A. Hero, “Dynamic metric learning from pairwise comparisons,” 54th Annual Allerton Conference on Communication, Control, and Computing, 2016 (invited). K. Greenewald, E. Zelnio, and A. Hero, “Robust SAR STAP via Kronecker decomposition,” IEEE Transactions on Aerospace and Electronic Systems, 2017. K. Moon, K. Sricharan, K. Greenewald, and A. Hero, “Improving Convergence of Divergence Functional Ensemble Estimators,” IEEE International Symposium on Information Theory, 2016. K. Greenewald, E. Zelnio, and A. Hero, “Kronecker STAP and SAR GMTI,” Proceedings of SPIE, 2016. K. Greenewald and A. Hero, “Robust Kronecker Product PCA for Spatio-Temporal Covariance Esti- mation,” IEEE Transactions on Signal Processing, 2015. K. Greenewald and A. Hero, “Regularized Block Toeplitz Covariance Matrix Estimation via Kronecker Product Expansions,” IEEE Workshop on Statistical Signal Processing (SSP), 2014 (invited). K. Greenewald and A. Hero, “Robust Kronecker Product PCA for Spatio-Temporal Covariance Estima- tion,” International Conference on Partial Least Squares and Related Methods (PLS), 2014 (invited). K. Greenewald and A. Hero, “Kronecker PCA based spatio-temporal modeling of video for dismount classification,” Proceedings of SPIE, 2014. K. Greenewald, T. Tsiligkaridis, and A. Hero, “Kronecker Sum Decompositions of Space-Time Data,” IEEE Workshop on Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP), Decem- ber 2013 (invited). K. Greenewald and B. Rigling, “Prediction of Optimal Bayesian Classification Performance for LADAR ATR,” MS Thesis, August 2012.
INSTRUCTION EXPERIENCE September-December 2016: Graduate Student Instructor, EECS 545. Primary graduate-level Machine Learning course. University of Michigan, Prof. Clayton Scott instructor.
PRESENTATIONS
“Information Flow, Neural Networks, and Generalization,” given at: Generality and Intelligence: from Biology to AI Workshop, Cambridge MA, October 2018. “Dynamic Metric Learning from Pairwise Comparisons,” given with variations at: University of Michigan, Graduate Student Statistical Topics Seminar Series, November 2016. University of Michigan, Michigan Student Symposium for Interdisciplinary Statistical Sciences, March 2016. MIT Lincoln Laboratory, Lexington, MA July 2015. “Robust Kronecker Product PCA for Spatio-Temporal Covariance Estimation,” given with variations at: University of Michigan, Graduate Student Statistical Topics Seminar Series, November 2014. University of Michigan, Michigan Student Symposium for Interdisciplinary Statistical Sciences, March 2015. “Kronecker PCA Based Space-Time Adaptive Processing,” given with variations at: Sensors Directorate, Air Force Research Laboratory, Dayton, OH, Aug. 2014. ATR Center Workshop, Wright State University, Dayton, OH, Aug. 2014. “Regularized Block Toeplitz Covariance Matrix Estimation via Kronecker Product Expansions,” given at: University of Michigan, Michigan Student Symposium for Interdisciplinary Statistical Sciences, March 2014. “Kronecker PCA for Spatio-Temporal Data,” given with variations at: Royal Observatory of Belgium, Brussels, Belgium, Nov. 2013. Ecole Superieur d’Electricite (Supelec), Gif-sur-Yvette, France, Nov. 2013. University of Michigan, Engineering Graduate Symposium, Nov. 2013. “Detection of Anomalous Crowd Behavior Using Spatio-Temporal Kronecker Sum Decompositions,” given with variations at: Sensors Directorate, Air Force Research Laboratory, Dayton, OH, Aug. 2013. ATR Center Workshop, Wright State University, Dayton, OH, Aug. 2013. “Prediction of Optimal Bayesian Classification Performance for LADAR ATR,” given at: Sensors Directorate, Air Force Research Laboratory, Dayton, OH, Aug. 2012.
SERVICE Michigan Student Symposium on Interdisciplinary Statistical Science (MSSISS), committee member, 2016. Organized campus-wide student symposium, with invited speaker Dr. Michael Jordan. Michigan Data Science Team, organizer, 2016. Spearheaded collaboration with MSSISS, helping orga- nize a student data challenge with monetary prizes. Reviewer for publications including ICML, COLT, NIPS (top 30% of reviewers: 2018), ICLR, AIS- TATS, UAI, IEEE ICASSP, IEEE CAMSAP, IEEE Workshop on Statistical Signal Processing, IEEE Transactions on Signal Processing, and IEEE Transactions on Image Processing.
OTHER
Programming Languages: Python, Matlab, C/C++.