Princeton University
Computer Science Department

Computer Science 598D
Boosting: Foundations & Algorithms

Rob Schapire

Spring 2009
Directory
General Information | Schedule & Readings | blackboard

Course Summary

We will read a draft of a forthcoming book on the boosting approach to machine learning.  Here is a current list of chapters, not all of which will be covered:
  1. Introduction and overview
  2. The mathematical study of machine learning
  3. Using AdaBoost to minimize the training error
  4. Direct bounds on the generalization error
  5. The margins explanation for boosting's effectiveness
  6. Game theory, on-line learning and boosting
  7. Loss minimization and generalizations of boosting
  8. Boosting, convex optimization and information geometry
  9. Consistency of boosting
  10. Optimal boosting and the continuous-time limit
  11. Improved boosting using confidence-rated predictions
  12. Multiclass classification problems
  13. Learning to rank

The course will be run as a reading group rather than a lecture class.

Administrative Information

Classes: 

Thursdays 1:30-4:20, CS Building, room 401.

Professor:

Rob Schapire:  407 CS Building, x8-7726, schapire@cs
Office hours: appointments can be made by visiting wass.princeton.edu (send me email if no convenient slots are available).

Prerequisites

COS 402 or 424 or 511, or permission of instructor.

No undergraduates, and no auditors or casual listeners.

The book

Boosting: Foundations and Algorithms
by Robert E. Schapire and Yoav Freund
to be published by MIT Press

Draft chapters from this book will be made available on blackboard (click on "course materials" then "readings").  You must be registered for the course to access these.  Because these are preliminary drafts, I am asking that you please not distribute them at all without my permission, and that you not keep an electronic copy on any computer, or place them where others can access them.

Note that the book is still incomplete in many regards.

Grading and workload

The class is mandatory P/D/F.  To get a passing grade (P), students are expected to do all readings, attend class regularly, and participate in discussions or otherwise provide feedback on what was read.  Students are also expected to lead discussions occasionally.

Communication

Once registered, you can send email to the entire class using blackboard (click on "communications" then "send email"), or you can send email directly to COS598D_S2009@ from your oit email account.