CSE 254 Spring 2005

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING
UNIVERSITY OF CALIFORNIA, SAN DIEGO

CSE 254: Seminar on Learning Algorithms

Spring 2005

CSE 254 is a graduate seminar devoted to recent research on AI learning methods and applications. This is not an introductory course, so the prerequisite is at least one graduate-level course (at UCSD or elsewhere) in machine learning or a closely related area such as statistics or pattern recognition. Appropriate courses at UCSD include CSE 291 (Statistical Learning), CSE 253 (Neural Networks for Pattern Recognition), and Cognitive Science 260 (Pattern Recognition).

The room for CSE 254 is Center Hall 201. The class meets on Tuesdays and Thursdays from 12:30 to 1:50. The first meeting will be on Tuesday March 29.

In each class meeting, a student will give a talk lasting about 60 minutes presenting a recent technical paper in detail. In questions during the talk, and in the final 20 minutes, all seminar participants will discuss the paper and the issues raised by it.


date	presenter	paper	author(s)	slides
March 29	organizational meeting
March 31	discussion of project guidelines
April 5	discussion on technical writing	Clear and Simple as the Truth (extracts)	Turner, Thomas
April 7	Daniel Hsu	Experiments with Random Projections for Machine Learning	Fradkin, Madigan	here
April 12	Shankar Shivappa	Theoretical Views of Boosting and Applications	Schapire	here
April 14	Paul Hammon	Feature selection, L1 vs. L2 regularization, and rotational invariance	Ng	here
April 19	Evan Ettinger	Grouping and dimensionality reduction by locally linear embedding	Perona, Polito	here
April 21	Charles Elkan	A new probabilistic model for documents
April 26	Shankar Shivappa	Recognition of Visual Speech Elements Using Adaptively Boosted Hidden Markov Models	Foo, Lian, Dong	here
April 28	Jan Voung	Online and Batch Learning of Pseudo-Metrics	Shalev-Shwartz, Singer, Ng	here
May 3	Nakul Verma	Algorithms for Large Scale Markov Blanket Discovery HITON, A Novel Markov Blanket Algorithm for Optimal Variable Selection	Tsamardinos, Aliferis, Statnikov	here
May 5	Mohsen Azarbayejani	Improving Text Classification by Shrinkage in a Hierarchy of Classes	McCallum, Rosenfeld, Mitchell, Ng	here
May 10	Jan Voung	Efficient Exact k-NN and Nonparametric Classification in High Dimensions	Liu, Moore, Gray	here
May 12	Evan Ettinger	Global versus local methods in nonlinear dimensionality reduction	de Silva, Tenenbaum	here
May 17	Charles Elkan	Yet another new probabilistic model for documents
May 19	Mohsen Azarbayejani	A Hierarchical Model for Clustering and Categorising Documents	Gaussier, Goutte, Popat, Chen
May 24	Doug Turnbull	Automatic music annotation	Turnbull	here
May 26	Shankar Shivappa	Exploiting generative models in discriminative classifiers An Information-Geometric Approach to Document Retrieval and Categorization	Jaakkola, Haussler Hofmann
May 31	Daniel Hsu	A random walks perspective on maximizing satisfaction and profit	Matthew Brand	here
June 2		Project presentations

Note: The seminar will run in parallel with a data mining contest sponsored by Fair Isaac, with cash prizes.

Each student will do one term project following specific guidelines. The project should be at the frontier of current research, and preferably closely inspired by at least one of the papers discussed in the class. Project reports will be evaluated using these grading criteria. There is a schedule for handing in a detailed project proposal, a draft project report, and then the final report.

The seminar will have no final exam. Letter grades will be based mostly on the final project report, but the presentations, participation in class and in the web-based discussions, and the intermediate project deliverables are all important also.

The instructor is Charles Elkan, Professor, whose office is AP&M room 4856. Feel free to send email to arrange an appointment, or telephone (858) 534-8897.

REGISTRATION

Students may take the seminar for a letter grade for four units, or for one or two units S/U:

For one unit, a student will present one research paper and participate in all class meetings.
For two units, a student will make two presentations.
Four units will require a presentation, participation in all class activities, and a project.

For four units, a student should register for CSE 254, section id 527966 for a letter grade. For one or two units, a student should register for the instructor's CSE 293, section id 527993.

Students who took a previous version of CSE 254 may take it again. Papers will be different this year.

PAPERS AND TOPICS

In the first week, we will make a schedule of papers and presentations for the whole quarter. Papers will be recent technical articles, often from NIPS and ICML. Each paper will be made available on the web as the quarter progresses. Students will choose papers in consultation with the instructor. Relevant topics may include:

supervised learning with many classes
regularization when observed counts are zero or small
discriminative versus generative modeling of data
semi-supervised learning from labeled and unlabeled data
transductive inference
new boosting algorithms
feature selection from very large feature spaces
modeling human heuristics for learning
reinforcement learning algorithms and applications
applications to text categorization
applications to image retrieval
applications in computational biology
financial applications

Some papers will be theoretical, and some will be applied. Each presentation will cover a single conference paper, to ensure that it is explained and discussed in sufficient depth.

Various textbooks are useful as background reading. including

Machine Learning by Tom Mitchell, McGraw Hill, 1997, ISBN 0070428077.
The Elements of Statistical Learning: Data Mining, Inference, and Prediction by T. Hastie, R. Tibshirani, and J. H. Friedman.

Students are encouraged to use other books and papers also.

PRESENTATIONS

The procedure for each student presentation is as follows:

One week in advance: Finish a draft of about 40 slides that present clearly the work in the paper. Make an appointment with the instructor to discuss the draft slides. Email the slides to elkan@cs.ucsd.edu.
Several days in advance: Meet for about one hour to discuss improving the slides, and how to give a good presentation.
Day of presentation: Give a good presentation with confidence, enthusiasm, and clarity.
Less than three days afterwards: Make changes to the slides suggested by the class discussion, and email the slides in PDF, two slides per page, to the instructor for publishing. Try to make your PDF file less than one megabyte.

Please read, reflect upon, and follow these presentation guidelines. Presentations will be evaluated, in a friendly way but with high standards, using this feedback form.

Each presentation should be prepared using LaTeX or Powerpoint, and should consist of about 40 slides. You must copy all important equations, diagrams, charts, and tables from the paper into your slides.

For each paper, we will have a web-based discussion area. Each student is expected to contribute at least one message to the discussion, before the presentation. A message may ask an interesting question, point out a strength or weakness of the paper, or answer a question asked by someone else. Messages should be thoughtful!

The schedule of presentations will be determined as much as possible on Tuesday March 29. Students should choose a date first, and then agree with the instructor about a paper to present. To find ideas, students can look at this list of possible papers and contact the instructor.

If you want to change your presentation date, please arrange a swap with another student and notify the instructor at least two weeks in advance.

Most recently updated on May 31, 2005 by Charles Elkan, elkan@cs.ucsd.edu