Steve Hanneke

Contact Information:
Email: steve.hanneke@gmail.com
Mobile Phone: (412) 973-3007
Location: Chicago, IL USA
TTIC Office: 507


I am a Research Assistant Professor at the Toyota Technological Institute at Chicago (TTIC).
I work on topics in statistical learning theory.

Research Interests:
My general research interest is in systems that can improve their performance with experience, a topic known as machine learning. My focus is on the statistical analysis of machine learning. The essential questions I am interested in answering are "what can be learned from empirical observation / experimentation," and "how much observation / experimentation is necessary and sufficient to learn it?" This overall topic intersects with several academic disciplines, including statistical learning theory, artificial intelligence, statistical inference, algorithmic and statistical information theories, probability theory, philosophy of science, and epistemology.

Brief Bio:
Prior to joining TTIC, I was an independent scientist working in Princeton 2012-2018, aside from a brief one-semester stint as a Visiting Lecturer at Princeton University in 2018. Before that, from 2009 to 2012, I was a Visiting Assistant Professor in the Department of Statistics at Carnegie Mellon University, also affiliated with the Machine Learning Department. I received my PhD in 2009 from the Machine Learning Department at Carnegie Mellon University, co-advised by Eric Xing and Larry Wasserman. My thesis work was on the theoretical foundations of active learning. From 2002 to 2005, I was an undergraduate studying Computer Science at the University of Illinois at Urbana-Champaign (UIUC), where I worked on semi-supervised learning with Prof. Dan Roth and the students in the Cognitive Computation Group. Prior to that, I studied Computer Science at Webster University in St. Louis, MO, where I played around with neural networks and classic AI a bit.

Recent News and Activities:

• New manuscript "A Theory of Universal Learning" posted to the arXiv.
• Received the Best Paper Award at COLT 2020 for our paper "Proper Learning, Helly Number, and an Optimal SVM Bound". [video]
• Presented (with Rob Nowak) an ICML 2019 Tutorial on Active Learning: From Theory to Practice. [video][slides]
• Organized the ALT 2019 workshop: When Smaller Sample Sizes Suffice for Learning.
• Fall 2018 I joined the Toyota Technological Institute at Chicago (TTIC) as a Research Assistant Professor.
• Spring 2018 I taught ORF 525 "Statistical Learning and Nonparametric Estimation" at Princeton University.
• My ICML 2007 paper "A Bound on the Label Complexity of Agnostic Active Learning" received Honorable Mention for the ICML 2017 Test of Time Award.
• Program Committee Chair (with Lev Reyzin) for the 28^th International Conference on Algorithmic Learning Theory (ALT 2017), held October 15-17 in Kyoto, Japan. See our published proceedings.
• New manuscript "Learning Whenever Learning is Possible: Universal Learning under General Stochastic Processes" posted to the arXiv.

Teaching:
Spring 2018: ORF 525, Statistical Learning and Nonparametric Estimation.
Spring 2012: 36-752, Advanced Probability Overview.
Fall 2011: 36-755, Advanced Statistical Theory I.
Spring 2011: 36-752, Advanced Probability Overview.
Fall 2010 Mini 1: 36-781, Advanced Statistical Methods I: Active Learning
Fall 2010 Mini 2: 36-782, Advanced Statistical Methods II: Advanced Topics in Machine Learning Theory
Spring 2010: 36-754, Advanced Probability II: Stochastic Processes.
Fall 2009: 36-752, Advanced Probability Overview.
At ALT 2010 and the 2010 Machine Learning Summer School in Canberra, Australia, I gave a tutorial on the theory of active learning. [slides]

A Survey of Theoretical Active Learning:

Theory of Active Learning. [pdf][ps]

This is a survey of some of the recent advances in the theory of active learning, with particular emphasis on label complexity guarantees for disagreement-based methods.
The current version (v1.1) was updated on September 22, 2014.
A few recent significant advances in active learning not yet covered in the survey: [ZC14], [WHE-Y15], [HY15].
An abbreviated version of this survey appeared in the Foundations and Trends in Machine Learning series, Volume 7, Issues 2-3, 2014.

Selected Recent Works:

Bousquet, O., Hanneke, S., Moran, S., van Handel, R., and Yehudayoff, A. (2020). A Theory of Universal Learning. Under Review.

Bousquet, O., Hanneke, S., Moran, S., and Zhivotovskiy, N. (2020). Proper Learning, Helly Number, and an Optimal SVM Bound. In Proceedings of the 33^rd Annual Conference on Learning Theory (COLT).

Montasser, O., Hanneke, S., and Srebro, N. (2019). VC Classes are Adversarially Robustly Learnable, but Only Improperly. In Proceedings of the 32^nd Annual Conference on Learning Theory (COLT).

Hanneke, S. (2020). Learning Whenever Learning is Possible: Universal Learning under General Stochastic Processes. Under Review.

Hanneke, S. (2016). The Optimal Sample Complexity of PAC Learning. Journal of Machine Learning Research, Vol. 17 (38), pp. 1-15.

Hanneke, S. and Yang, L. (2015). Minimax Analysis of Active Learning. Journal of Machine Learning Research, Vol. 16 (12), pp. 3487-3602.

Hanneke, S. (2012). Activized Learning: Transforming Passive to Active with Improved Label Complexity. Journal of Machine Learning Research, Vol. 13 (5), pp. 1469-1587.

Articles in Preparation:

Nonparametric Active Learning, Part 1: Smooth Regression Functions. [pdf][ps].

Nonparametric Active Learning, Part 2: Smooth Decision Boundaries.

Learning Whenever Learning is Possible: Universal Learning under General Stochastic Processes. [pdf][ps][arXiv].

Active Learning with Identifiable Mixture Models. Joint work with Vittorio Castelli and Liu Yang.

A No-Free-Lunch Theorem for MultiTask Learning. Joint work with Samory Kpotufe. [pdf][arXiv].

Stable Sample Compression Schemes: New Applications and an Optimal SVM Margin Bound. Joint work with Aryeh Kontorovich. [pdf][arXiv].

A Theory of Universal Learning. Joint work with Olivier Bousquet, Shay Moran, Ramon van Handel, and Amir Yehudayoff. [pdf][arXiv].

All Publications: (authors are listed in alphabetical order, except sometimes when a student author is listed first).

2020

Hanneke, S., Kontorovich, A., Sabato, S., and Weiss, R. (2020). Universal Bayes Consistency in Metric Spaces. The Annals of Statistics, (to appear). [pdf][arXiv].

Montasser, O., Hanneke, S. and Srebro, N. (2020). Reducing Adversarially Robust Learning to Non-Robust PAC Learning. Advances in Neural Information Processing Systems 34 (NeurIPS). [pdf][arXiv].

Bousquet, O., Hanneke, S., Moran, S., and Zhivotovskiy, N. (2020). Proper Learning, Helly Number, and an Optimal SVM Bound. In Proceedings of the 33^rd Annual Conference on Learning Theory (COLT). [pdf][official page][arXiv][video].
Winner of the Best Paper Award.

2019

Hanneke, S. and Yang, L. (2019). Surrogate Losses in Passive and Active Learning. Electronic Journal of Statistics, Vol. 13 (2), pp. 4646-4708. [pdf][ps][journal page][arXiv].

Hanneke, S. and Kpotufe, S. (2019). On the Value of Target Data in Transfer Learning. Advances in Neural Information Processing Systems 33 (NeurIPS). [pdf][official page][arXiv].

Hanneke, S. and Kontorovich, A. (2019). Optimality of SVM: Novel Proofs and Tighter Bounds. Theoretical Computer Science. Volume 796, Pages 99-113. [pdf][journal page]

Montasser, O., Hanneke, S., and Srebro, N. (2019). VC Classes are Adversarially Robustly Learnable, but Only Improperly. In Proceedings of the 32^nd Annual Conference on Learning Theory (COLT). [pdf][official page][arXiv]
Winner of a Best Student Paper Award.

Hanneke, S. and Kontorovich, A. (2019). A Sharp Lower Bound for Agnostic Learning with Sample Compression Schemes. In Proceedings of the 30^th International Conference on Algorithmic Learning Theory (ALT). [pdf][official page][arXiv]

Hanneke, S., Kontorovich, A., and Sadigurschi, M. (2019). Sample Compression for Real-Valued Learners. In Proceedings of the 30^th International Conference on Algorithmic Learning Theory (ALT). [pdf][official page][arXiv]

Hanneke, S. and Yang, L. (2019). Statistical Learning under Nonstationary Mixing Processes. In Proceedings of the 22^nd International Conference on Artificial Intelligence and Statistics (AISTATS). [pdf][official page][arXiv]

2018

Hanneke, S. and Yang, L. (2018). Testing Piecewise Functions. Theoretical Computer Science, Vol. 745, pp. 23-35. [pdf][ps][journal page][arXiv]

Zhivotovskiy, N. and Hanneke, S. (2018). Localization of VC Classes: Beyond Local Rademacher Complexities. Theoretical Computer Science, Vol. 742, pp. 27-49. [pdf][ps][journal page][arXiv]
(Special Issue for ALT 2016; Invited)

Hanneke, S., Kalai, A., Kamath, G., and Tzamos, C. (2018). Actively Avoiding Nonsense in Generative Models. In Proceedings of the 31^st Annual Conference on Learning Theory (COLT). [pdf][official page][arXiv]

Yang, L., Hanneke, S., and Carbonell, J. (2018). Bounds on the Minimax Rate for Estimating a Prior over a VC Class from Independent Learning Tasks. Theoretical Computer Science, Vol. 716, pp. 124-140. [pdf][ps][journal page][arXiv]
(Special Issue for ALT 2015; Invited)

2016

Zhivotovskiy, N. and Hanneke, S. (2016). Localization of VC Classes: Beyond Local Rademacher Complexities. In Proceedings of the 27^th International Conference on Algorithmic Learning Theory (ALT). [pdf][ps][arXiv]

Hanneke, S. (2016). Refined Error Bounds for Several Learning Algorithms. Journal of Machine Learning Research, Vol. 17 (135), pp. 1-55. [pdf][ps][arXiv][journal page]

Hanneke, S. (2016). The Optimal Sample Complexity of PAC Learning. Journal of Machine Learning Research, Vol. 17 (38), pp. 1-15. [pdf][ps][arXiv][journal page]

2015

Hanneke, S. and Yang, L. (2015). Minimax Analysis of Active Learning. Journal of Machine Learning Research, Vol. 16 (12), pp. 3487-3602. [pdf][ps][arXiv][journal page]

Hanneke, S., Kanade, V., and Yang, L. (2015). Learning with a Drifting Target Concept. In Proceedings of the 26^th International Conference on Algorithmic Learning Theory (ALT). [pdf][ps][arXiv]
See also this note on a result for the sample complexity of efficient agnostic learning implicit in the above concept drift paper: [pdf]

Yang, L., Hanneke, S., and Carbonell, J. (2015). Bounds on the Minimax Rate for Estimating a Prior over a VC Class from Independent Learning Tasks. In Proceedings of the 26^th International Conference on Algorithmic Learning Theory (ALT). [pdf][ps][arXiv]

Wiener, Y., Hanneke, S., and El-Yaniv, R. (2015). A Compression Technique for Analyzing Disagreement-Based Active Learning. Journal of Machine Learning Research, Vol. 16 (4), pp. 713-745. [pdf][ps][arXiv][journal page]

2014

Hanneke, S. (2014). Theory of Disagreement-Based Active Learning. Foundations and Trends in Machine Learning, Vol. 7 (2-3), pp. 131-309. [official] [Amazon] 
There is also an extended version, which I update from time to time.

2013

Yang, L. and Hanneke, S. (2013). Activized Learning with Uniform Classification Noise. In Proceedings of the 30^th International Conference on Machine Learning (ICML). [pdf][ps][appendix pdf][appendix ps]

Yang, L., Hanneke, S., and Carbonell, J. (2013). A Theory of Transfer Learning with Applications to Active Learning. Machine Learning, Vol. 90 (2), pp. 161-189. [pdf][ps][journal page]

2012

Balcan, M.-F. and Hanneke, S. (2012). Robust Interactive Learning. In Proceedings of the 25^th Annual Conference on Learning Theory (COLT).[pdf][ps][arXiv]

Hanneke, S. (2012). Activized Learning: Transforming Passive to Active with Improved Label Complexity. Journal of Machine Learning Research, Vol. 13 (5), pp. 1469-1587. [pdf][ps][arXiv][journal page]
Related material: extended abstract, Chapter 4 in my thesis, and various presentations [slides][video].

2011

Yang, L., Hanneke, S., and Carbonell, J. (2011). Identifiability of Priors from Bounded Sample Sizes with Applications to Transfer Learning. In Proceedings of the 24^th Annual Conference on Learning Theory (COLT).[pdf][ps][video]

Yang, L., Hanneke, S., and Carbonell, J. (2011). The Sample Complexity of Self-Verifying Bayesian Active Learning. In Proceedings of the 14^th International Conference on Artificial Intelligence and Statistics (AISTATS).[pdf][ps]

Hanneke, S. (2011). Rates of Convergence in Active Learning. The Annals of Statistics, Vol. 39 (1), pp. 333-361. [pdf][ps][journal page]

2010

Yang, L., Hanneke, S., and Carbonell, J. (2010). Bayesian Active Learning Using Arbitrary Binary Valued Queries. In Proceedings of the 21^st International Conference on Algorithmic Learning Theory (ALT).[pdf][ps]
Also available in information theory jargon. [pdf][ps]

Hanneke, S., Fu, W., and Xing, E.P. (2010). Discrete Temporal Models of Social Networks. The Electronic Journal of Statistics, Vol. 4, pp. 585-605. [pdf][journal page]

Hanneke, S. and Yang, L. (2010). Negative Results for Active Learning with Convex Losses. Proceedings of the 13^th International Conference on Artificial Intelligence and Statistics (AISTATS). [pdf][ps]

Balcan, M.-F., Hanneke, S., and Wortman Vaughan, J. (2010). The True Sample Complexity of Active Learning. Machine Learning, Vol. 80 (2-3), pp. 111-139. [pdf][ps][journal page]
(Special Issue for COLT 2008; Invited)

2009

Hanneke, S. (2009). Theoretical Foundations of Active Learning. Doctoral Dissertation. Machine Learning Department. Carnegie Mellon University. [pdf][ps][defense slides]

Hanneke, S. (2009). Adaptive Rates of Convergence in Active Learning. In Proceedings of the 22^nd Annual Conference on Learning Theory (COLT).[pdf][ps][slides]
Also available in expanded journal version.

Hanneke, S. and Xing, E.P. (2009). Network Completion and Survey Sampling. In Proceedings of the 12^th International Conference on Artificial Intelligence and Statistics (AISTATS).[pdf][ps][slides]

2008

Balcan, M.-F., Hanneke, S., and Wortman, J. (2008). The True Sample Complexity of Active Learning. In Proceedings of the 21^st Annual Conference on Learning Theory (COLT). [pdf][ps][slides]
Winner of the Mark Fulk Best Student Paper Award.
Also available in an extended journal version.

2007

Balcan, M.-F., Even-Dar, E., Hanneke, S., Kearns, M., Mansour, Y., and Wortman, J. (2007). Asymptotic Active Learning. NIPS Workshop on Principles of Learning Problem Design. [pdf][ps][spotlight slide]
Also available in improved conference version and expanded journal version.

Hanneke, S. and Xing, E.P. (2007). Network Completion and Survey Sampling. NIPS Workshop on Statistical Network Models.
See our later conference publication.

Hanneke, S. (2007). Teaching Dimension and the Complexity of Active Learning. In proceedings of the 20^th Annual Conference on Learning Theory (COLT). [pdf][ps][slides]

Hanneke, S. (2007). A Bound on the Label Complexity of Agnostic Active Learning. In proceedings of the 24^th Annual International Conference on Machine Learning (ICML). [pdf][ps][slides]
Honorable Mention for the ICML 2017 Test of Time Award.

Guo, F., Hanneke, S., Fu, W., and Xing, E.P. (2007). Recovering Temporally Rewiring Networks: A Model-based Approach. In proceedings of the 24^th Annual International Conference on Machine Learning (ICML). [pdf]
Also see our related earlier work.

Hanneke, S. (2007). The Complexity of Interactive Machine Learning. KDD Project Report (aka Master's Thesis). Machine Learning Department, Carnegie Mellon University. [pdf][ps][slides]
Includes some interesting results from a class project on The Cost Complexity of Interactive Learning, in addition to my COLT07 and ICML07 papers.

2006

Hanneke, S. and Xing, E.P. (2006). Discrete Temporal Models of Social Networks. In Proceedings of the ICML Workshop on Statistical Network Analysis. [pdf][ps][slides]
Also available in an extended journal version

Hanneke, S. (2006). An Analysis of Graph Cut Size for Transductive Learning. In Proceedings of the 23^rd International Conference on Machine Learning (ICML). [pdf][ps][slides ppt][slides pdf]