NUIM CS401 F2005
Mon 12:00-12:50 CS2
Tue 15:00-15:50 Hume6
Instructor: Barak
Pearlmutter, barak@cs.nuim.ie
Office: Hamilton Institute (NUIM,
Rye Hall, South Wing, room 5)
Office hours: you are welcome any time, just drop on by. (Afternoons
are best, excepting every third Friday.) Or feel free to email or
ring me up (x6394) and make an appointment.
Text
Machine Learning, by Tom Mitchell.
Lectures
- (19-Sep-2005)
Intro to Machine Learning
- (20-Sep-2005)
Perceptron Learning Rule (ch4 1/4)
- (26-Sep-2005)
Linear and semilinear units (ch4 2/4)
- (27-Sep-2005)
Convergence of LMS (ch4 3/4)
- (3-Oct-2005)
Derivation of Backpropagation
- (4-Oct-2005)
Lab: implementation of perceptron, backpropagation
- (10-Oct-2005)
Generalization curves (ch4 4/4)
- (11-Oct-2005)
Information Theory,
Bayes' Rule,
KL-Divergence,
Entropy
- (17-Oct-2005)
More Bayes' Rule,
Evidence, Prior,
Gaussian Distributions,
Hypothesis Space,
weight decay
- (18-Oct-2005)
EM of mixture of iid binary vectors
- (24-Oct-2005)
EM of mixture of Gaussians
- (25-Oct-2005)
HMMs
- (7-Nov-2005)
HMMs part duex
- (8-Nov-2005)
HMMs part shalosh (w/ some practice)
- (14-Nov-2005)
Support Vector Machines
- (15-Nov-2005)
Support Vector Machines II
- (21-Nov-2005)
Decision Trees I
- (22-Nov-2005)
Decision Trees II
- (28-Nov-2005)
VC Dimension
- (5-Dec-2005)
PAC learning; strong learning vs weak learning;
boosting and the ADAboost algorithm
- (6-Dec-2005)
Multiplicative updates and irrelevant features (WINNOW)
- (12-Dec-2005)
Review of Assignment 3; rants about modularity
- (13-Dec-2005)
Review
Useful Materials
Author's lecture notes from the textbook: TM-book
Code written in class, usually scrubbed up a bit.
Notes
HMM Cheat Sheet
Assignments
Pertinent answers to questions about the assignments will be added to
the bottom of the assignment pages. Please check there before asking
a question, as it might already be answered. But if it is not, please
do ask - this will be of benefit not only to you but to everyone in
the class.
- (19-Sep-2005) machine learning application brainstorm
- (4-Oct-2005) fun with perceptrons and backprop
- (28-Nov-2005) Battle HMM!
- (13-Dec-2005) take home practice quiz
Assignment-Related Warnings
- Late assignments will not be accepted.
- Don't cheat OR
ELSE.
(Talking about the assignments, and even getting some debugging help
from your friends, is fine. Please give them credit though: "Thanks
to Sally Helperman for assistance finding a mysterious bug in blah
blah blah, and to Sam Sucker for spending seven hours showing me how
to use Excel to generate ugly plots." And obviously copying, or not
actually doing the work yourself, is not okay.)
- Sometimes I'll talk to people about what they turned in for an
assignment. If you can't discuss your work cogently with me, I might
be led to suspect that it was not in fact your own work.
- Even though assignments are only 30% of the grade, successful
completion of the course is contingent upon satisfactory performance
on the assignments. (In other words: if you don't do the assignments
you don't pass the course even if you score 100% on exams.)