Suriya Gunasekar

[email protected]
http://sgunasekar.github.io/
July 24, 2019

Current position
2016–Present Research Assistant Professor.
Toyota Technological Institute at Chicago, USA.
Mentor: Nathan Srebro.
Primary research topics: (a) inductive bias from optimization in learning and (b) supervised learn-
ing of non–discriminatory predictors in the context of algorithmic fairness.
Summer 2019 Long term visitor.
Simons Institute for the Theory of Computing, University of California, Berkeley.
Special program on Foundations of Deep Learning..

Education
2010–2016 MS—PhD in Electrical and Computer Engineering.
The University of Texas at Austin, USA.
Advisor: Prof. Joydeep Ghosh.
PhD Thesis: Mining Structured Matrices in High Dimensions
2006–2010 B. Tech in Electronics and Communication Engineering.
National Institute of Technology, Warangal, India.

Grants, awards, and select scholarships

2018 Senior Research Personnel on National Science Foundation award AF: RI: Medium: Collaborative
Research: Understanding and Improving Optimization for Deep and Recurrent Networks.
(was involved in writing and submission of the proposal)
2018 Top 200 (10%) reviewer for Conference on Neural Information Processing Systems (NeurIPS).
2006-2010 Institute Merit Scholarship from NIT Warangal for four academic years 2006–2010
(awarded annually to top 10% of the meritorious students at the institute).
2009 INSA-IASc-NASI Summer Research Fellowship from Indian National Science Academy
(awarded based on selection of short term research proposals).

Teaching
2018 Co-instructor. Introduction to Machine Learning Summer School jointly organized by TTI Chicago
and the University of Chicago as part of an NSF Research and Training Grant.
• Two week intensive course on basics of machine learning primarily aimed at graduate students from
outside of computer science departments.
• Covered all topics of a full-term ML course. Teaching material at http://www.ttic.edu/intromlss2018.
2017–2018 Lectures at TTI Japan on Introduction to Machine Learning.
• Taught 4 lectures in the remote course jointly offered by multiple TTI Chicago faculty (Spring ‘17, ‘18).
• Topics: Classification, AdaBoost, generative models, mixture models, Expectation-Maximization (EM).
2016–2017 Guest lectures at TTI Chicago on Introduction to Statistical Machine Learning.
• Total of 4 lectures during Fall ‘16, ‘17 (Instructor: Dr. Greg Shaknarovich).
• Topics: Regularization, gradient descent, logistic regression, support vector machines, kernels.

http://sgunasekar.github.io/ • [email protected] 1/5

2006–2010 Teaching Assistant, UT Austin.
Data Mining, Spring ‘13 (Instructor: Dr. Joydeep Ghosh).
Digital Logic Design, Fall ‘10 – Fall ‘11 (Instructors: Dr. Adnan Aziz, Dr. Lizy John).

Publications
Peer-reviewed full length conference publications
1. Lexicographic and Depth-Sensitive Margins in Homogeneous and Non-Homogeneous Deep Mod-
els. M. S. Nacson, S. Gunasekar, J. Lee, N. Srebro, D. Soudry. In 36th International Conference on
Machine Learning (ICML) 2019.
2. Convergence of gradient descent on separable data. M. S. Nacson, J. Lee, S. Gunasekar, P. Savarese,
N. Srebro, D. Soudry. In 22nd International Conference on Artifical Intelligence and Statistics (AIS-
TATS) 2019.
3. Implicit bias of gradient descent on linear convolutional networks. S. Gunasekar, J. Lee, D. Soudry,
N. Srebro. In 32nd Conference on Neural Information Processing Systems (NeurIPS) 2018.
4. On preserving non–discrimination when combining expert advice. A. Blum, S. Gunasekar, T. Lyk-
ouris, N. Srebro. In 32nd Conference on Neural Information Processing Systems (NeurIPS) 2018.
5. Characterizing implicit bias in terms of optimization geometry. S. Gunasekar, J. Lee, D. Soudry, N.
Srebro. In 35th International Conference on Machine Learning (ICML) 2018.
6. Implicit regularization in matrix factorization. S. Gunasekar, B. Woodworth, S. Bhojanapalli, B.
Neyshabur, N. Srebro. In 31st Conference on Neural Information Processing Systems (NIPS) 2017.
(spotlight presentation)
7. Learning non–discriminatory predictors. B. Woodworth, S. Gunasekar, M. Ohannessian, N. Srebro.
In 30th Conference on Learning Theory (COLT) 2017.
8. Preference completion from partial rankings. S. Gunasekar, O. Koyejo, J. Ghosh. In 30th Conference
on Neural Information Processing Systems (NIPS) 2016.
9. Identifiable phenotyping using constrained non–negative matrix factorization. S. Joshi, S. Gu-
nasekar, D. Sontag, J. Ghosh. In 1st Machine Learning for Healthcare Conference (MLHC) 2016.
10. Unified view of matrix completion under general structural constraints. S. Gunasekar, A. Banerjee,
J. Ghosh. In 29th Conference on Neural Information Processing Systems (NIPS) 2015.
11. Consistent collective matrix completion under joint low rank structure. S. Gunasekar, M. Yamada,
D. Yin, Y. Chang. In 18th International Conference on Artificial Intelligence and Statistics 2015.
12. Face detection on distorted images using perceptual quality–aware features. S. Gunasekar, J.
Ghosh, A. Bovik. In 14th IS&T/SPIE Human Vision and Electronic Imaging Conference 2014.
13. Exponential family matrix completion under structural constraints. S. Gunasekar, P. Ravikumar,
J. Ghosh. In 31st International Conference on Machine Learning (ICML) 2014.
14. Noisy matrix completion using alternating minimization. S. Gunasekar, A. Acharya, N. Gaur, J.
Ghosh. In Machine Learning and Knowledge Discovery in Databases (ECML/PKDD) 2013.
15. Review quality aware collaborative filtering. S. Raghavan, S. Gunasekar, J. Ghosh. In 6th ACM
Conference on Recommender Systems (RecSys) 2012.

Journal publications
16. The implicit bias of gradient descent on separable data. D. Soudry, E. Hoffer, M. S. Nacson, S.
Gunasekar, N. Srebro. In Journal of Machine Learning Research 2018.
17. Face detection on distorted images augmented by perceptual quality–aware features. S. Gunasekar,
J. Ghosh, A. Bovik. In IEEE Transactions on Information Forensics and Security 2014.

http://sgunasekar.github.io/ • [email protected] 2/5

 Preprints/Theses
• Kernel and Deep Regimes in Overparametrized Models. B. Woodworth, S. Gunasekar, J. Lee, D.
Soudry, N. Srebro. In Arxiv 2019.
• Phenotyping using structured collective matrix factorization of multi-source EHR data. S. Gu-
nasekar, J. Ho, J. Ghosh, S. Kreml, A. N. Kho, J. C. Denny, B. A. Malin, J. Sun. In Arxiv 2016.
Dissertation Mining structured matrices in high dimensions.
Advised by Prof. Joydeep Ghosh. In UT Electronic Theses and Dissertations, 2016.
Master’s A survey on using side information in recommendation systems.
Report Advised by Prof. Joydeep Ghosh. In UT Electronic Theses and Dissertations, 2012.

Invited talks/seminars
2019 Kernel and deep regimes in overlarameterized learning at the workshop on Frontiers of Deep Learn-
ing at the Simons Institute, University of California, Berkeley.
2019 Characterizing Optimization Bias in terms of Optimization Geometry at Data Science through a
Geometric Lens workshop, Symposium of Theory of Computing (STOC).
2019 Rethinking the Role of Optimization in Learning at
• University of British Columbia, Canada,
• University of Illinois Urbana-Champaign,
• University of Washington at Seattle,
• University of California at San Diego,
• New York University,
• McGill University, Canada,
• Ecole polytechnique federale de Lausanne (EPFL), Switzerland,
• ETH Zurich, Switzerland,
• Duke University,
• University of Wisconsin Madison,
• University of Massachusetts at Amherst,
• Johns Hopkins University,
• University of Michigan at Ann Arbor,
• Columbia University,
• Princeton University,
• Georgia Institute of Technology,
• Microsoft Research, Redmond,
• Northwestern University.
2019 Optimization bias in linear convolutional networks at MIFODS workshop on Non-convex optimiza-
tion and deep learning, MIT.
2018 Implicit Bias of Optimization in Learning at
• INFORMS Annual Meeting,
• Center for Applied Mathematics (CAM) colloquium, Cornell University,
• International Symposium on Mathematical Programming (ISMP),
• Microsoft Research, Redmond.
2018 Optimization Geometry and Implicit Regularization at SILO Seminar, UW Madison.
2017 Implicit Regularization in Matrix Factorization at
• Statistics colloquium, Indiana University Bloomington,
• RIKEN Center for Advanced Intelligent Project (AIP), Japan,
• Information Theory and Applications (ITA) workshop.
2017 Regularization in non-convex Optimization at CSL/SINE Seminar, UIUC.

http://sgunasekar.github.io/ • [email protected] 3/5

 2016 Mining Structured Matrices in High Dimensions at
• Toyota Technological Institute at Chicago,
• Google Research, New York.

Conference presentations
Oral presentations
2017 Characterizing implicit bias in terms of optimization geometry at International Conference on Ma-
chine Learning (ICML).
2016 Implicit regularization in matrix factorization (spotlight presentation) at Conference on Neural In-
formation Processing Systems (NIPS).
2014 Exponential family matrix completion under structural constraints. at International Conference on
Machine Learning (ICML).
2012 Review Quality Aware Collaborative Filtering at Conference on Recommender Systems (RecSys).

Poster presentations
2018 Implicit bias of gradient descent on linear convolutional networks at NIPS.
2016 Preference completion from partial rankings at NIPS.
2015 Unified view of matrix completion under general structural constraints at NIPS.
2014 Face detection on distorted images using perceptual quality–aware features at IS&T/SPIE Human
Vision and Electronic Imaging Conference.

Professional activities
Organization/Committees
2018 Organizer and co-instructor for the Introduction to Machine Learning Summer School jointly
organized by TTIC and UChicago. Website: http://ttic.edu/intromlss2018
2018 Organizing committee member for the 2nd Midwest Machine Learning Symposium (MMLS).
Website: http://midwest-ml.org/2018/
2018 Committee member on drafting of the Policy on Discrimination, Harassment, and Abusive Behav-
ior at TTIC.
2016-Present Organizer for Machine Learning Seminar Series at TTIC. Website: http://www.ttic.edu/mls/.

Journal reviewing
Journal of Machine Learning Research, Mathematics of Operations Research, IEEE Transactions
on Knowledge and Data Engineering, IEEE Signal Processing Letters, IEEE Transactions on Signal
Processing, IEEE Transactions on Image Processing, Neurocomputing Journal.

Conference reviewing
Senior PC/ Neural Information Processing Systems (NeurIPS), 2019; Uncertainty in Artificial Intelligence
area chair (UAI) 2020.
PC/reviewer International Conference on Machine Learning (ICML) , Neural Information Processing Systems
(NeurIPS), International Conference on Learning Representations (ICLR), Uncertainty in Artificial
Intelligence (UAI), Conference on Learning Theory (COLT), Algorithmic Learning Theory (ALT),
Association for the Advancement of Artificial Intelligence Conference (AAAI), Conference on Data
Sciences (CODS), International Joint Conference on Artificial Intelligence (IJCAI).
• Outstanding reviewer for ICML 2018.
• Top 200 (10%) reviewer for NeurIPS 2018.

http://sgunasekar.github.io/ • [email protected] 4/5

 Past employment
2010–2016 Graduate Research Assistant. The University of Texas Austin, USA.
Worked on research questions within multiple NSF funded projects.
Supervisors: Dr. Joydeep Ghosh and Dr. Alan C. Bovik.
• Listwise ranking in high dimensions using retargeting of preference feedback.
• Interpretable phenotype extraction from patient electronic health records (EHRs).
• Robust face detection models using image quality indicative features.
Summer ‘14 Research Intern. Yahoo Labs, Sunnyvale, USA.
Worked on theoretical analysis and applications of collective matrix completion.
Summer ‘12 Research Intern. SRI Labs, Princeton, USA.
Worked on models for personalized photo recommendation system.
Summer ‘11 Software Intern. Apple Inc., Cupertino, USA.
Worked on data collection and preliminary data analysis prototypes for hardware testing.

References
1. Joydeep Ghosh, Professor, Department of ECE, [email protected]
The University of Texas at Austin.
2. Nathan Srebro, Professor, [email protected]
Toyota Technological Institute at Chicago.
3. Avrim Blum, Professor and Chief Academic Officer, [email protected]
Toyota Technological Institute at Chicago.
4. Robert Nowak, Professor, Department of ECE, [email protected]
University of Wisconsin, Madison.

http://sgunasekar.github.io/ • [email protected] 5/5