Difference between revisions of "10-601 Ensembles"
From Cohen Courses
Jump to navigationJump to search| (5 intermediate revisions by the same user not shown) | |||
| Line 1: | Line 1: | ||
| − | This a lecture used in the [[Syllabus for Machine Learning 10- | + | This a lecture used in the [[Syllabus for Machine Learning 10-601B in Spring 2016]] |
=== Slides === | === Slides === | ||
| − | * | + | * William's [http://www.cs.cmu.edu/~wcohen/10-601/ensembles.ppt Slides in PowerPoint], [http://www.cs.cmu.edu/~wcohen/10-601/ensembles.pdf Slides in PDF]. |
| − | + | * Margin "movie" William showed in class: [http://www.cs.cmu.edu/~wcohen/10-601/margin-movie.pdf Margin movie]. | |
| − | * Margin "movie" | ||
=== Readings === | === Readings === | ||
| + | * Murphy 16.4, 16.6 | ||
| + | |||
| + | Other optional readings: | ||
* [http://dl.acm.org/citation.cfm?id=743935 Ensemble Methods in Machine Learning], Tom Dietterich | * [http://dl.acm.org/citation.cfm?id=743935 Ensemble Methods in Machine Learning], Tom Dietterich | ||
* [http://cseweb.ucsd.edu/~yfreund/papers/IntroToBoosting.pdf A Short Introduction to Boosting], Yoav Freund and Robert Schapire. | * [http://cseweb.ucsd.edu/~yfreund/papers/IntroToBoosting.pdf A Short Introduction to Boosting], Yoav Freund and Robert Schapire. | ||
Latest revision as of 15:45, 6 January 2016
This a lecture used in the Syllabus for Machine Learning 10-601B in Spring 2016
Slides
- William's Slides in PowerPoint, Slides in PDF.
- Margin "movie" William showed in class: Margin movie.
Readings
- Murphy 16.4, 16.6
Other optional readings:
- Ensemble Methods in Machine Learning, Tom Dietterich
- A Short Introduction to Boosting, Yoav Freund and Robert Schapire.
- Optional: Improved boosting algorithms using confidence-rated predictions, Robert Schapire and Yoram Singer. (This paper has the analysis that I presented in class.)
Summary
You should understand how these ensemble methods work
- Bagging
- Stacking
- Boosting
You should understand the basic intuitions behind the analysis of boosting:
- As reducing an upper bound on error and hence fitting the training data.
- As a coordinate descent optimization of the same upper bound.
You should also be aware that boosting is related to margin classifiers.
