CMPS 290C Home Page
Advanced Machine Learning
An Introduction to Proabilistic Graphical Models Spring 2004

Manfred K. Warmuth


Fifteen minute project presentations, We June 9th, BE 318, noon-2pm

Projects due on Th, June 10th, 6pm.
Pls slide printout under my office (if possible) and link report
and talk into the files proj/proj.html

Put both your project and talk file into that directory

CLASS PROJECTS


Organisational 
       Class:	TTh 4-5:45, Porter 241
Office hours:	Mo,We 10-11, 331 Baskin Engineering
Prerequisite:	CMPS 242 - Machine Learning 
		or a graduate Bayesian Statistics class
Textbook by Michael I. Jordan: Same as title of course.
Will hand out hard copies of chapters as we go along

Good tutorial material 

	Friedman-Goldszmidt

Summary of lectures 

1:	General introduction
	Introductory article by Jordan 
	Introductory article by Friedman on CompBio Apps 

2:	Tutorial by Friedman on
I-maps, d-seperation, minimal I-maps


3:	Sets of independences implied by digraphs 

4:	Sets of independences implied by undirected graphs 
	Elimination algorithm
	Homework 1, Due Tu 4-20-04 in class

5: 	Sum product algorithm
        Factor graphs

6:	Computing max. a posteriori probabilities
        Formulations for general semi-rings
        Interpolating between maximum and sum
	Free energy
	Application to speech
	Semi-ring version of all pairs shortest path algorithm
        Bayesian versus Frequentists

7:	Regression - mixtures - classification
	Solutions, Homework 1

8:	LMS, LLS, Steepest Descent, 
	QR decomposition, Pseudo-inverse, SVD
	Ridge regression, shrinkage
	Homework 2, Due Tu 5-4-04 in class

9:	Linear classification, generative versus
	discriminative, Logistic Regression

10:	Newton's method, IRLS for Logistic Regression,
	trigger variable, noisy or
	Exponential families of distributions

11:	GLIM, sufficient statistics, ML, KL to empirical distribution
	Alternate to GLIM based on two Bregman divergences
	PartII of on-line learning tutorial
        
	One page project proposal, due Tu, May 11

	Solutions homework 3

	Homework 3, Due Th 5-27-04 in class

12:	Summary of GLIM 
	Divergence based approach for deriving updates
	Matching loss function
	Completely observed graphical models

13:	Mixtures, conditional mixtures
	Various derivations of EM
	Implicit versus explicit updates

14:	HMMs as graphical model
	Joint Entropy Updates for Gaussian Mixtures
	Joint Entropy Updates HMMs

15:	Multivariate Gaussians
	Factor analysis

16:	Kalman filtering

17: 	Morkov properties

18:	Junction tree algorithm
	Solutions, Homework 3

19:	Junction tree algorithm (continued)
        HMMs as special case

20:	Building and exponential model starting with features
     	Maxent connection
	Generalization to Bregman divergences
	
	Learning a positive definite matrix/kernel
	Extended abstract
 
        Open problems



----------------------------------------------------------
Last modified Wednesday, 05-Apr-2000 23:13:17 PDT