6.438 Algorithms for Inference
Fall 2011
Carl Friedrich Gauss and Andrey Andreyevich Markov
Instructors: Devavrat Shah, Gregory W Wornell
TAs: George H Chen, Roger Baker Grosse, George Jay Tucker
Lecture: TR9.30-11 (32-124)
Information:
Introduction to statistical inference with probabilistic
graphical models. Directed and undirected graphical models; factor
graphs; Gaussian models. Hidden Markov models, linear dynamical
systems. Sum‐product and junction tree algorithm. Forward‐backward
algorithm; Kalman filtering and smoothing. Variational methods,
mean‐field theory, and loopy belief propagation. Particle methods
and filtering. Min‐sum algorithm; Viterbi algorithm. Building
graphical models from data; parameter estimation, learning
structure. Selected special topics.
ACCESS TO THE WEBSITE is limited to the students enrolled in the
course as LISTENERS or FOR CREDIT. Please talk to us in class about
registering and getting access to the web site. You can download
the general information handout by clicking on the "General
Info" tab on the left, even if you are not enrolled in the
course.
Announcements
quiz, solutions, and histogram posted
Hi all,The second quiz has been posted on Stellar, along with the solutions and histogram.
Roger
Announced on 15 December 2011 9:55 p.m. by Roger Baker Grosse
all notes posted; reminders
Hi all,1. By now, all lecture and recitation notes, as well as all recitation videos, have been posted to Stellar. (There are no videos for the final recitation.)
2. As a reminder, the final quiz is tomorrow, Monday Dec. 12, 7-10pm in 32-155. You're allowed 4 8.5/11" sheets of notes, both sides.
3. As a reminder, there are no office hours tomorrow from 5-7pm, but there are office hours 11am-12 in the usual place.
4. A bug in the likelihood expression for Gaussian mixture models in the recitation 11 notes has been fixed.
Roger
Announced on 11 December 2011 10:22 p.m. by Roger Baker Grosse
Homework 9 graded
Hi all,Homework 9 has been graded. You can pick it up either at recitation tomorrow or at office hours.
Roger
Announced on 08 December 2011 11:18 p.m. by Roger Baker Grosse
Office hours this week
Hi all,Because of the upcoming quiz, we're shifting the TA office hour schedule a bit. The new schedule is as follows:
Friday, 3-5pm
Sunday, 3-4pm
Monday, 11-12am
Wednesday 4-5pm is still available by appointment.
Roger
Announced on 06 December 2011 12:27 a.m. by Roger Baker Grosse
Rec. 11 notes are up: many remarks on the EM algorithm, full coverage of Baum-Welch, etc
Hello all.I've gotten many questions regarding the last two lectures. I'll try to put up lecture notes some time later this weekend as well as some other resources that will hopefully fill in details.
For now, I've put up (extended!) recitation 11 notes, which includes lots of remarks about the EM algorithm and full coverage of the Baum-Welch algorithm (i.e. EM for HMM's) as well as the Gaussian mixture model example that I didn't have time to get to. Hopefully these examples help solidify how EM works.
Those of you who are using the alpha-beta version of the forward-backward algorithm for Problem 9.2 and want to know how to compute edge marginals using alpha-beta messages, check out equation (8) in the recitation notes (includes derivation, which uses facts from Lecture 9 notes relating alpha-beta messages to BP messages).
Meanwhile, several students asked about a Bayesian version of EM (i.e. for computing MAP rather than ML estimates), so I've included a brief discussion of this at the end of the recitation notes as well.
Cheers,
George
Announced on 03 December 2011 5:45 p.m. by George H Chen
