Structure learning: why?
Lecture 15
Model selection/ structure learning
Koller & Friedman chapter 14
Mackay chapter 28
We often want to learn the structure of the graphical model:
Scientic discovery (data mining)
Use a good model for prediction, compre
Generative vs discriminative models
Generative model denes P (h, o) = P (h)P (o|h).
HMM (hidden Markov model)
Lecture 14
P (ht|ht1)][
P (h, o) = [
t
MRF (Markov random eld)
1
P (h, o) = [
Z
Kevin Murphy
p(ot|ht)]
t
i j Ni
H2
H1
3 November 2004
O1
P (oi
MLE for general Bayes nets
(K+F 13.113.2, J 9.19.2)
Lecture 13:
Parameter learning in undirected models
If we assume the parameters for each CPD are globally
independent, then the log-likelihood function decomposes into a
sum of local terms, one per node
MLE for general Bayes nets
Lecture 11:
Bayesian Parameter Learning
If we assume the parameters for each CPD are globally independent,
and all nodes are fully observed, then the log-likelihood function
decomposes into a sum of local terms, one per node:
p
Administrivia
Probabilistic graphical models
CPSC 532c (Topics in AI)
Stat 521a (Topics in multivariate analysis)
HW4 due today
Lecture 9
Kevin Murphy
Monday 18 October 2004
Review
Belief propagation
Variable elimination can be used to answer a single q
Administrivia
Probabilistic graphical models
CPSC 532c (Topics in AI)
Stat 521a (Topics in multivariate analysis)
Next Monday: no class (thanksgiving)
Next Wednesday: lecture by Brent Boerlage.
Lecture 8
Kevin Murphy
Wednesday 6 October, 2004
Recall var
Administrivia
Probabilistic graphical models
CPSC 532c (Topics in AI)
Stat 521a (Topics in multivariate analysis)
Homework 2 is now due on Wednesday 29th.
Please start reading chapters 6 and 7 before Wednesday.
Lecture 5
Kevin Murphy
Monday 27 September
Administrivia
Probabilistic graphical models
CPSC 532c (Topics in AI)
Stat 521a (Topics in multivariate analysis)
Mark Crowley will hold a regular discussion section on Fridays 1-2pm,
CICSR 304. He will discuss HW1 and give a Matlab tutorial in the
rst m
Administrivia
Probabilistic graphical models
CPSC 532c (Topics in AI)
Stat 521a (Topics in multivariate analysis)
Spare stapled copies of the book chapters are outside my door (107).
If you take the last unstapled copy, please photocopy and return to
the
Administrivia
Probabilistic graphical models
CPSC 532c (Topics in AI)
Stat 521a (Topics in multivariate analysis)
Lecture 2
Class web page http:/www.cs.ubc.ca/murphyk
/Teaching/CS532c Fall04/index.html
Send email to [email protected] with the content