6.438 Algorithms for Inference
Fall 2013
Instructor: Devavrat Shah
TAs: Qingqing Huang, Georgios Papachristoudis
Lecture:
TR9.30-11
(32-124)
Recitation: F10-11, 11-12
(36-372)
Office hours (Devavrat): R11-12 (by appt)
(32-D670)
Office hours (Qingqing): M2pm-3pm
(32-D6th floor lounge)
Office hours (Giorgos): M6:30pm-7:45pm
(32-D4th floor lounge)
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.
Announcements
Class project grades out
Dear Inference Gurus,We have finished evaluating everyone's project reports and we are stunned by your terrific performance! Class, as a whole, has done above and beyond what we had expected and for that we would like to congratulate all of you!
We are providing the summary of everyone's performance. The first chart (pareto_results.png) shows your performance in terms of normalized log-likelihood and time (sort_norm_score.pdf). The second pdf file shows your normalized scores and times (in an anonymized list). Highlights from project descriptions will be uploaded later in the day. On the Gradebook module, you can check your grade (from 1-5), which took into account the completeness of your report, the novelty of your method, the ability of your code to run and its performance.
Finally, it is our pleasure to announce the five winners who
have captured the "pareto boundary" of the performance
curve. They are (in alphabetical order) :
Chong Yang Goh (cygoh)
William Lotter (lotter)
Eric Shyu (eshyu)
Guolong Su (guolong)
Jennifer Tang (jstang)
Again, terrific performance and we thank you for that.
Best wishes with the final exam!
Giorgos, Qingqing, Devavrat
ps: norm_score is computed as: norm_score = 0.5*mean(norm_loglik) + 0.5*(1-norm_time). norm_loglik for each part (1, 2, 3) is computed as: norm_loglik = loglik - min(loglik); norm_loglik = norm_loglik/max(norm_loglik);, so that it captures the performance of someone relative to all other students. Similarly, for norm_time.
Announced on 10 December 2013 1:59 p.m. by Georgios Papachristoudis
Final exam today
Dear class,As a friendly reminder, the final exam takes place today (12/10)
at 7pm at 32-155 room!
A few more hours and you will be released from 6.438's tyranny
or as a Greek saying puts it, "Now that you ate a whole
donkey, only the tail has been left!"
Good luck for today!
ps: Disclaimer, we do not eat donkeys in Greece!
Announced on 10 December 2013 10:13 a.m. by Georgios Papachristoudis
Wrapping up the final project
6.438 sailors,I hope you have either already submitted your project or you're one step before submitting. To ease you a little bit, we're going to leave the submission link open a few more hours in case you want to re-upload something or in case you delayed a little bit.
Keep Calm and Think of Holidays!
Announced on 03 December 2013 1:12 a.m. by Georgios Papachristoudis
Online Subject Evaluations
Class please take a moment to evaluate the class and us!http://web.mit.edu/subjectevaluation
* Comments are greatly appreciated!
* Your feedback is read and valued.
Announced on 02 December 2013 6:08 p.m. by Georgios Papachristoudis
Scores of Makeup Quiz
Class,You can see your score in the Gradebook module under the name
"Make Up Quiz".
In addition, you can see you final score out of the two quizzes in
"Max(Quiz1,Makeup)"
Best.
Announced on 14 November 2013 9:07 a.m. by Georgios Papachristoudis
