Soumya Ray

(Ph. D., University of Wisconsin-Madison, 2005)
Assistant Professor
Department of Electrical Engineering and Computer Science
Case Western Reserve University
Office: Olin 516
Email: sray AT case_edu
Mailing Address: Department of EECS, Glennan 320, 10900 Euclid Ave, Cleveland OH 44106-7071
Research Areas: Artificial Intelligence, Machine Learning, Reinforcement Learning and Planning

Teaching

I teach undergraduate AI (EECS 391) in the spring and graduate machine learning (EECS 440) in the fall. I have also co-taught graduate AI (EECS 491) in the fall; however, this is being split into two courses starting Fall 2013. I also co-ordinate the EECS 500 colloquium series each semester. You can visit this page to learn more about these courses.

I am the current AI minor advisor.

Current teaching: Spring 2013: EECS 391 (Introduction to Artificial Intelligence)

If you are interested in machine learning research, here is my reading group:Machine Learning Reading Group. You can send me email if you want to know what we are reading. Please note that you will need to have taken EECS 440 or 491 or equivalent to contribute effectively.

Students

Gary Doran (Ph.D.)
Kai Liang (MS)
Nathan McKinley (MS)
Scott Sosnowski (MS)
Tyler Goeringer (MS)
Feng Cao (MS, graduated Spring 2012, at Amazon inc.)
Thesis: Classification, Detection and Prediction of Adverse and Anomalous Events in Medical Robots
Tim Ernsberger (MS, graduated Fall 2012, at Amazon inc.)
Thesis: A Hierarchical Integration of Planning and Reinforcement Learning
Howie Richmond (MS, graduated Fall 2011, at MIM Software, co-supervised with Andy Podgurski)
Thesis: Knowledge Transfer for Incremental Fault Localization

Current Research

Bayesian Hierarchical Reinforcement Learning
Humans manage complex tasks by decomposition and also using prior information collected by solving other tasks. How can we integrate these ideas into autonomous agents? This research seeks ways to answer this question. (Students: Kai Liang, Feng Cao)
Kernel Methods for Multiple-Instance Learning
Kernels are powerful and flexible representation transformations of feature vectors which enable nonlinear classifiers to be learned efficiently in a supervised setting. How do these behave when applied to multiple-instance data, where sets of feature vectors are mapped? A variety of surprising behaviors emerge in this case. This research seeks to characterize this behavior and understand the strengths and weaknesses of this technique for MI data. (Students: Gary Doran)
Machine Learning for Software Engineering
The scale and complexity of modern software makes it prone to bugs. Automated techniques using machine learning can assist developers during testing and debugging to quickly locate and remove faults. This research seeks to develop automated and collaborative methods to detects software defects and improve software reliability. With Andy Podgurski. (Students: Boya Sun, Gang Shu, Zhuofu Bai, Howie Richmond)
Machine Learning for Automated Stent Detection in OCT images
Millions of people receive stent implants as treatment for coronary artery disease. Subsequently, these are imaged to see if further intervention is needed if complications arise. These images are currently manually processed by radiologists and take 6-16 hours per stent in intravascular Optical Coherence Tomography (iOCT) images. We are developing automated techniques using machine learning that can effectively help radiologists process these images in an hour or less, reducing cost and lowering errors. With David Wilson. (Students: Hong Lu, CWRU Biomedical Engineering)
Preventing Adverse and Anomalous Events in Cyber-physical systems
Many modern robotic systems are used in situations where reliability is critical, but how to estimate reliability of these complex systems under various circumstances is not well understood. We are designing processes to estimate the reliability of these systems, in particular to detect and prevent adverse and anomalous (A&A) events in medical robots online. With Andy Podgurski and Cenk Cavusoglu. Supported by NSF. (Students: Kai Liang, Feng Cao, Zhuofu Bai, Mark Renfrew)

Previous Research

Machine Learning for spam filtering
We are working on adaptive methods for spam detection. Our goal is to develop methods that can detect these messages as early as possible in network traffic, thereby saving bandwidth and decreasing congestion. With Michael Rabinovich and Mark Allman. (Students: Tu Ouyang)
Knowledge Transfer in Reinforcement Learning
"Transfer Learning" focuses on methods that can effectively transfer knowledge acquired about one task to help in solving another, different task. We are developing techniques that transfer knowledge between different sequential decision processes, using real-time strategy games as our testbeds. With Alan Fern, Prasad Tadepalli, Tom Dietterich. (Students: Neville Mehta and Aaron Wilson)
Efficient learning for hard Boolean functions
Certain Boolean functions, such as parity, are known to be hard to learn efficiently. In this work, we demonstrated that the hardness of learning these functions is linked to the input distribution of the data; if the input distribution is "significantly different" from the uniform distribution, these functions may be efficiently learnable. Based on this observation, we developed a method called Skewing that is often able to learn such functions efficiently, given enough observations. With David Page, Lisa Hellerstein, Bard Rosell, Eric Lantz and Eric Bach.
Learning from Multiple-Instance Data
In standard supervised learning, examples are described by a tuple of attribute-value pairs. In some problems, such as predicting the binding affinities of small molecules to a target protein, examples are described by sets of such tuples. We have developed new algorithms for classification and regression from such "multiple-instance" data, and shown that the straightforward extension of linear regression to this setting is NP-complete. With Mark Craven, David Page and Burr Settles.
Information Extraction from Free Text
Information extraction is the task of creating structured relations out of free text. In our work, we have developed statistical methods for doing this that also incorporate grammatical information about sentences obtained using an automated parser, Sundance. With Mark Craven and Marios Skounakis.
Machine Learning for Question Answering
Question answering systems are designed to provide accurate responses to short factual questions asked in natural language. In our work, we have developed a method that the system can use to learn from past questions to improve accuracy on future questions. With Eric Brill. (This work is not publicly available)

Publications

G. Doran and S. Ray (2013).
SMILe: Shuffled Multiple-Instance Learning.
To appear in the Proceedings of the 27th AAAI Conference on Artificial Intelligence, Bellevue, Washington, USA.
K. Liang, F. Cao, Z. Bai, M. Renfrew, M. Cenk Cavusoglu, A. Podgurski and S. Ray (2013).
Detection and Prediction of Adverse and Anomalous Events in Medical Robots.
To appear in the Proceedings of the 25th Annual Conference on Innovative Applications of Artificial Intelligence (IAAI), Bellevue, Washington, USA.
S. Sosnowski, T. Ernsberger, F. Cao and S. Ray (2013).
SEPIA: A Scalable Game Environment for Artificial Intelligence Teaching and Research.
To appear in the Proceedings of the Fourth Symposium on Educational Advances in Artificial Intelligence (EAAI), Bellevue, Washington, USA.
H. Lu, M. Gargesha, Z. Wang, D. Chamie, G. F. Attizzani, T. Kanaya, S. Ray, M. A. Costa, A. M. Rollins, H. G. Bezerra and D. L. Wilson (2013).
Automatic stent strut detection in intravascular OCT images using image processing and classification technique.
Appears in the Proceedings of SPIE Medical Imaging: Computer-Aided Diagnosis, vol 8670, eds. Carol Novak, Stephen Aylward.
F. Cao and S. Ray (2012).
Bayesian Hierarchical Reinforcement Learning.
Appears in the Proceedings of the 26th Annual Conference on Neural Information Processing Systems (NIPS), Lake Tahoe, Nevada, USA.
H. Lu, M. Gargesha, Z. Wang, D. Chamie, G. F. Attizzani, T. Kanaya, S. Ray, M. A. Costa, A. M. Rollins, H. G. Bezerra and D. L. Wilson (2012).
Automatic stent detection in intravascular OCT images using bagged decision trees.
Biomedical Optics Express, vol 3 no. 11(Nov):2809--2824.
M. Ruffalo, M. Koyuturk, S. Ray and T. LaFramboise (2012).
Accurate Estimation of Short Read Mapping Quality for Next Generation Genome Sequencing.
Appears in the Proceedings of the 11th European Conference on Computational Biology, Basel, Switzerland. (also published in a special issue in Bioinformatics)
B. Sun, G. Shu, A. Podgurski and S. Ray (2012).
CARIAL: Cost-Aware Software Reliability Improvement with Active Learning.
In Proceedings of the Fifth IEEE International Conference on Software Testing, Verification and Validation (ICST), Montreal, CA.
N. Mehta, S. Ray, P. Tadepalli and T. Dietterich (2011).
Automatic Discovery and Transfer of Task Hierarchies in Reinforcement Learning.
Appears in AI Magazine, special issue on Transfer of Structured Knowledge. (reviewed article)
T. Ouyang, S. Ray, M. Rabinovich and M. Allman (2011).
Can Network Characteristics Detect Spam Effectively in a Stand-Alone Enterprise?
In Proceedings of the 12th Passive and Active Measurement Conference, Atlanta, GA, USA.
B. Sun, A. Podgurski and S. Ray (2010).
Improving the Precision of Dependence-Based Defect Mining by Supervised Learning of Rule and Violation Graphs.
In Proceedings of the 21st IEEE International Symposium on Software Reliability Engineering, San Jose, CA, USA.
L. Hellerstein, B. Rosell, E. Bach, S. Ray and D. Page (2009).
Exploiting Product Distributions to Identify Relevant Variables of Correlation Immune Functions. Journal of Machine Learning Research, vol 10(Oct):2374--2411. Tech report version
Eric Bach's paper on improved bounds for the number of correlation-immune functions.
N. Mehta, S. Ray, P. Tadepalli and T. Dietterich (2008).
Automatic Discovery and Transfer of MAXQ Hierarchies. pdf
Appears in the Proceedings of the 25th International Conference on Machine Learning, Helsinki, Finland.
B. Settles, M. Craven and S. Ray (2007).
Multiple-Instance Active Learning. pdf
Appears in the Proceedings of the 21st Conference on Neural Information Processing Systems, Vancouver, BC, Canada.
H. Chan, A. Fern, S. Ray, N. Wilson and C. Ventura (2007).
Online Planning for Resource Production in Real-Time Strategy Games. pdf
Appears in the Proceedings of the 17th International Conference on Automated Planning & Scheduling, Providence, RI, USA.
E. Lantz, S. Ray and D. Page (2007).
Learning Bayesian Network Structure from Correlation-Immune Data. pdf
Appears in the Proceedings of the 23rd Conference on Uncertainty in Artificial Intelligence, Vancouver, BC, Canada.
A. Wilson, A. Fern, S. Ray and P. Tadepalli (2007).
Multi-task Reinforcement Learning: A Hierarchical Bayesian Approach. pdf ps.gz
Appears in the Proceedings of the 24th International Conference on Machine Learning, Corvallis, OR, USA.
J. Davis, V. S. Costa, S. Ray and D. Page (2007).
An Integrated Approach to Feature Invention and Model Construction for Drug Activity Prediction. pdf ps.gz
Appears in the Proceedings of the 24th International Conference on Machine Learning, Corvallis, OR, USA.
S. Ray (2005).
Learning from Data with Complex Interactions and Ambiguous Labels. ps pdf ps.gz
PhD thesis, Department of Computer Sciences, University of Wisconsin-Madison, Madison, WI, USA.
S. Ray & M. Craven (2005).
Supervised versus Multiple-Instance Learning: An Empirical Comparison. ps pdf ps.gz
Appears in the Proceedings of the 22nd International Conference on Machine Learning, Bonn, Germany.
B. Rosell, L. Hellerstein, S. Ray & D. Page (2005).
Why Skewing works: Learning Difficult Boolean Functions with Greedy Tree Learners. ps pdf ps.gz
Appears in the Proceedings of the 22nd International Conference on Machine Learning, Bonn, Germany.
S. Ray & D. Page (2005).
Generalized Skewing for Functions with Continuous and Nominal Attributes. ps pdf ps.gz
Appears in the Proceedings of the 22nd International Conference on Machine Learning, Bonn, Germany.
S. Ray & M. Craven (2005).
Learning Statistical Models for Annotating Proteins with Function Information using Biomedical Text.
Appears in BMC Bioinformatics, Vol 6(Suppl 1). online ps pdf ps.gz
S. Ray & D. Page (2004).
Sequential Skewing: An Improved Skewing Algorithm. ps pdf ps.gz
Appears in the Proceedings of the 21st International Conference on Machine Learning, Banff, Canada.
D. Page & S. Ray (2003).
Skewing: An Efficient Alternative to Lookahead for Decision Tree Induction. ps pdf ps.gz
Appears in the Proceedings of the 18th International Joint Conference on Artificial Intelligence, Acapulco, Mexico.
M. Skounakis, M. Craven & S. Ray (2003).
Hierarchical Hidden Markov Models for Information Extraction. pdf
Appears in the Proceedings of the 18th International Joint Conference on Artificial Intelligence, Acapulco, Mexico.

S. Ray & M. Craven (2001).
Representing Sentence Structure in Hidden Markov Models for Information Extraction. ps pdf ps.gz
Appears in the Proceedings of the 17th International Joint Conference on Artificial Intelligence, Seattle, WA, USA.

S. Ray & D. Page (2001).
Multiple Instance Regression. ps pdf ps.gz
Appears in the Proceedings of the 18th International Conference on Machine Learning, Williamstown, MA, USA.

Workshop Publications

A. Wilson, A. Fern, S. Ray, P. Tadepalli (2008).
Learning and Transferring Roles in Multi-Agent MDPs. pdf
Transfer Learning for Complex Tasks Workshop, 23rd AAAI Conference on Artificial Intelligence, Chicago, USA.
N. Mehta, M. Wynkoop, S. Ray, P. Tadepalli and T. Dietterich (2007).
Automatic Induction of MAXQ Hierarchies. pdf
Hierarchical Organization of Behavior Workshop, 21st Conference on Neural Information Processing Systems, Vancouver, BC, Canada.
H. Chan, A. Fern, S. Ray, N. Wilson and C. Ventura (2007).
Extending Online Planning for Resource Production in Real-Time Strategy Games with Search. pdf
Workshop on Planning in Games, ICAPS 2007, Providence, RI, USA.

Contributed Chapters

S. Ray, S. Scott and H. Blockeel (2009). Multiple Instance Learning.
Encyclopedia of Machine Learning, eds C. Sammut and G. Webb. Springer. ISBN: 978-0-387-30768-8, Springer.
S. Ray and P. Tadepalli (2009). Model-based Reinforcement Learning.
Encyclopedia of Machine Learning,eds C. Sammut and G. Webb. Springer. ISBN: 978-0-387-30768-8, Springer.

Posters and Technical Reports

G. Doran and S. Ray (2012). Kernel methods for Multiple-Instance Learning. Poster at the 29th International Conference on Machine Learning, Edinburgh, Scotland.
C-Y. Wu, S. Ray, K. P. Barry and A. I. Jack (2010). On Functional Annotation of the Human Brain by Combining Resting State Connectivity and Activation Foci. In the {\em 40th Annual Meeting of the Society for Neuroscience}, San Diego, CA, USA (poster).
T. Ouyang, S. Ray, M. Allman and M. Rabinovich (2009). A Large Scale Empirical Analysis of Email Spam Detection Through Transport Level Characteristics. Technical Report TR 10-001, International Computer Science Institute, CA, USA.

Miscellaneous Activities

I serve on the Case School of Engineering's graduate studies committee.
I recently served on the Program Committee of ICML 2013 and the Senior program committee of AAAI 2013.
I am currently serving on the editorial board of Machine Learning journal.
I was Registration Chair for the International Conference on Machine Learning (ICML) 2007 held at Oregon State University .
I was the webmaster for the website of the International Machine Learning Society till December 2008. The new webmaster is Sebastian Stark.
I am the current co-moderator for the ML-news Google group.

Information Extraction Datasets

Literature and Datasets for the Multiple Instance Problem

Machine Learning News

Visit this group

Home IIT Kharagpur UW-Computer Sciences Oregon State EECS Case Western EECS

Last update 5/16/2013 by Soumya Ray