CS678 - Spring 2002
Cornell University
Department of Computer Science

• Monday, 1:25pm - 2:15pm in Hollister Hall 110
• Wednesday, 1:25pm - 2:15pm in Hollister Hall 110
• Friday, 1:25pm - 2:15pm in Hollister Hall 110

• Rich Caruana, caruana@cs.cornell.edu, 4155 Upson Hall, office hours Tuesday 14:00-15:00, Wednesday 14:30-15:00
• Thorsten Joachims, tj@cs.cornell.edu, 4153 Upson Hall, office hours Monday 14:30-15:00, Thursday 14:30-15:00, Friday 14:30-15:00

1. SVM Clustering:
   - B. Sch�lkopf, J. Platt, J. Shawe-Taylor, A. J. Smola, and R. C. Williamson. Estimating the support of a high-dimensional distribution. Technical Report 99-87, Microsoft Research, 1999. To appear in Neural Computation, 2001.
   - Ben-Hur et al., Support Vector Clustering. JMLR, 2, 2001.
   (1-2 students, 20 minutes, April 22/24)
2. SVM Regression:
   - A. J. Smola and B. Sch�lkopf. A tutorial on support vector regression. NeuroCOLT Technical Report NC-TR-98-030, Royal Holloway College, University of London, UK, 1998. To appear in Statistics and Computing, 2001. (pages 1-14 only)
   (1 student, 20 minutes, Feb. 27 or March 1)
3. Kernel Principal Component Analysis:
   Bernhard Sch�lkopf, Alexander Smola, Klaus-Robert M�ller, Kernel Principal Component Analysis, in:  B. Scholkopf, C. Burges, and A. Smola, editors, Advances in Kernel Methods --- Support Vector Learning. MIT Press, Cambridge, MA, 1999. 327 -- 352. Short version or chapter in Support Vector Learning  for background.
   (1 student, 20 minutes, April 12/15/17)
4. Multi-Class SVMs:
   - John Platt, Large-Margin DAGs for Multi-Class Classification, NIPS 2000.
   (1 student, 20 minutes, Feb. 27 or March 1)
5. Learning Rankings:
   - William W. Cohen, Robert E. Schapire, Yoram Singer, Learning to order things, Journal of Artificial Intelligence Research, 10, 1999.
   (1 student, 20 minutes, April 1/3).
6. Boosting/Bagging:
   - Leo Breiman, Arcing Classifiers, Machine Learning, 1998.
   - Eric Bauer, Roni Kohavi, An Empirical Comparison of Voting Classification Algorithms: Bagging, Boosting, and Variants, Machine Learning, 1999.
   (2 students, 30 minutes, Feb. 22/25/27)
7. Learning to Learn:
   - Sebastian Thrun and Joseph O'Sullivan, Discovering Structure in Multiple Learning Tasks: The TC Algorithm, ICML-96.
   (1 student, 20 minutes, March 6/8)
8. Clustering:
   - P.S. Bradley, Usama Fayyad, and Cory Reina, Scaling Clustering Algorithms to Large Databases, AAAI-98.
   - P.S. Bradley and Usama Fayyad, Refining Initial Points for K-Means Clustering, ICML-98.
   (1 student, 30 minutes, April 29 or May 1)
9. Graphical Models:
   -
   (1 student, 20 minutes, March 15/25/27)
10. ROC and Related Methods:
    -Foster Provost and Tom Fawcett, Analysis and Visualization of Classifier Performance: Comparison under Imprecise Class and Cost Distributions, KDD-97 (ROC Convex Hull Method)
    -Chris Drummond and Robert C. Holte, Explicitly Representing Expected Cost: An Alternative to ROC Representation, KDD-2000.
    (1 student, 25 minutes, April 19/22/24)

• Support Vector Machines and Kernel-based Methods: VC theory, optimal hyperplane and maximum-margin separation, soft-margin, SVMs for regression, Mercer kernels, error bounds, leave-one-out bounds,  quadratic programming, connections to related methods (8 lectures)
• Unsupervised Learning and Clustering: agglomerative clustering, distributional clustering, k-means, Bayesian clustering, principal component analysis, scaling issues for large datasets (6 lectures)
• Bayes Nets: inference, maximum likelihood estimation, latent variables, expectation/maximization, hidden Markov models, learning structure, causality (5 lectures)
• Boosting and Bagging: Adaboost, bias/variance, margins (5 lectures)
• Error Estimation and Model Selection: no free lunch, bias/variance, Bayesian learning, minimum description length, Leave-one-out and cross-validation, holdout testing, bootstrap estimation (3 lectures)
• Learning to Order Data: learning retrieval functions in information retrieval, learning for ROC analysis  (3 lectures)
• Inductive Transfer: Learning multiple related tasks (3  lectures)
• Reinforcement Learning: Markov decision processes, finite state models, Q-learning, dynamic programming (1-2 lecture)

We will illustrate methods and theory with practical examples in the areas of information retrieval, language technology, and medical decision making.

• Duda, Hart, Stork, "Pattern Classification"
• Devroye, Gyoerfi, Lugosi, "A Probabilistic Theory of Pattern Recognition"
• Shawe-Taylor, Cristianini, "Introduction to Support Vector Machines"
• Hastie, Tibshirani, Friedman, "The Elements of Statistical Learning"
• Vapnik, "Statistical Learning Theory"
• Sutton, Barto, "Reinforcement Learning"
• Mitchell, "Machine Learning"

• CS478
• CS578
• equivalent of any of the above
• permission from the instructors

• 20%: Homework: (4 homeworks max, some programming, some non-programming)
• 20%: MidTerm Exam (take home)
• 20%: Final Exam (in class)
• 20%: Student Paper Presentations
• 20%: Final Projects

Roughly: A=90-100; B=80-90; C=70-80; D=60-70; F= below 60