This course will be offered again in Spring 2002.

Students taking 254 for four units should follow these project guidelines.  We now have a feedback form for presentations.  On Wednesday June 4, each person doing a project gave a short talk about his or her project.  Here are the final project reports.

Please read, reflect upon, and follow these presentation guidelines.  Immediately after your presentation, please email to elkan@cs a copy of your slides.  For ease of viewing, please make this copy be two slides per page in Adobe PDF.

The schedule of papers and presentations is below.  Second presentations are on June 4 and later.  Participants who do not yet have a presentation date may "bump" these second presentations.  Participants who have not chosen a paper yet should 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.
 

OVERVIEW

Students may take CSE 254 for one, two, or four units:

• For one unit, a student will present one paper as described below.
• For two units, a student will make two separate presentations.  (This option may be eliminated.)
• Four units will require one or two presentations and a project.

CSE 254 is not offered every academic year, so interested students should enroll this quarter.  CSE 254 is open to all students in computer science, cognitive science, or engineering who have already taken at least one graduate course at UCSD or elsewhere in artificial intelligence or a closely related area such as statistics, pattern recognition, bioinformatics, or information theory.  Appropriate courses at UCSD include CSE 250A, CSE 253, Cognitive Science 202, ECE  270A, and CSE 275A.

The seminar will meet on Mondays and Wednesdays from 10:10am to 11:30pm in the APM building, room 5218 (not 3218 as announced previously).  The first meeting will be on Monday April 2, and the final meeting will be on Wednesday June 6, 2001.

Topics to be covered in CSE 254 may include:

• estimating well-calibrated probabilities
• estimating probabilities when observed counts are zero or small
• discriminative versus generative modeling of data
• semi-supervised learning from labeled and unlabeled data
• transductive inference
• bias-variance decomposition analysis of bagging and boosting
• new boosting algorithms
• feature selection via boosting
• support vector methods for learning with more features than examples
• new adaptive methods for regularization and overfitting prevention
• learning when the training and test sets have different probability distributions
• modeling human heuristics for learning
• new reinforcement learning applications
• applications to text categorization
• applications to image retrieval
• applications to gene array data and (speculative!) protein folding
• applications to financial prediction

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

One textbook is recommended as background reading for CSE 254: Machine Learning by Tom Mitchell, McGraw Hill, 1997, ISBN 0070428077.
 
 

SEMINAR ORGANIZATION

Some papers will be theoretical, and some will be applied.  Two related applications papers may be discussed together.  Theoretical papers will typically be presented and discussed alone, to ensure that mathematical and algorithmic questions are discussed in sufficient depth.

In the first week, we will make a schedule of papers and presentations for the whole quarter.  With 10 participants, each student will make two separate presentations.  The procedure for one presentation is as follows:

• Two weeks in advance: Finish a draft of 25 to 30 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.
• Ten days in advance: Meet for 30 to 60 minutes to discuss improving the slides, and how to give a good presentation.
• Day of presentation: Give a good presentation with confidence, enthusiasm, and clarity.
• Immediately afterwards: Make changes to the slides suggested by the class discussion.  Email the slides in PDF, two slides per page, to the instructor for publishing.

For each presentation, we will have a web-based discussion area.  Each seminar participant 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!

Each student will also do one term project following specific guidelines.  The project should be at the frontier of current research, and preferably closely inspired by 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.  Final grades will be based 50% on presentations and participation in class and in the web-based discussions.  The other 50% will be the project report.
 
 

Most recently updated on June 16, 2001 by Charles Elkan, elkan@cs.ucsd.edu