Supplemental Readings

• "Bagging, boosting, and C4.5", by J.R. Quinlan, AAAI, 1996
• "Experiments with a new boosting algorithm", by Y. Freund and R.E. Schapire, ICML, 1996
• Multi-class Adaboost (SAMME) by J. Zhu, S. Rosset, H. Zou and T. Hastie, Technical Report, Department of Statistics, University of Michigan, 2006
• Efficient multiclass boosting classification with active learning (GAMBLE) by J. Huang, S. Ertekin, Y. Song, H. Zha, and C. Lee Giles, SDM 2007

• An MCMC applet
• A tutorial
• The original paper by Hastings published in Biometrika (1970).
• Sequential Monte Carlo Methods portal (UK mirror)
• Papers and tutorials on MCMC by Peter Green and collaborators in the Statistics Group of University of Bristol, UK

• A Java-based software package ideal for quick-and-easy construction and exploration of Bayes nets. (Here are some tips for use. Launch the program by going into "Classes" directory and typing "java JavaBayes". Two windows will come up. In the console window select File->Open. Navigate to Examples directory and pick dog-problem.xml or any other example network.)
• An extensive collection of links to graphical models software packages compiled by Kevin Murphy.

• A widely-used (except for PA2!) implementation of SVM. Scroll down the page to play with a real-time SVM classification applet.
• Another widely-used (but also not eligibile for PA2) SVM implementation.
• An advanced SVM tutorial by Chris Burges.

• Gaussian Process basics videolecture presented by DavidMacKay.
• A Gaussian Processes "portal" with links to books, software, and cutting-edge research results.

• A kernels "portal" with links to books, software, and cutting-edge research results.

• Here is a motivating example for probit regression. It comes from Jim Albert
and his book on Bayesian computation with "R".

In Programming Assignent 1 we have explored Fisher's linear discriminant to tell apart faces from non-faces. By contrast, face recognition means attaching a unique label (e.g., Lena or Lenin) to an image of a face. A technique by P.N. Belhumer, J.P. Hespanha, and D.J. Kriegman, "Eigenfaces vs. Fisherfaces: Recognition Using Class Specific Linear Projection", cleverly uses ideas from Fisher LDA for this multi-class classification problem.

• Thomas Bayes died a long time ago, but his ideas are not just the stuff of the textbooks. Why not make a pot of hot cider and curl up with a philosophically deep and mathematically rigourous treatise on Bayesian Statistics by José M. Bernardo?
• Do u totally dig math? Then you'll like this introduction to probability by Rich Bass.
• Can we model the entire world with only a Bayes rule? We'll ask you this question at the end of the course. But take a look at one noteworthy extension known as Dempster-Shafer theory and summarized by Glenn Shafer himself.