Introductory Applied Machine Learning
Course Homepage
Lecturers: Victor Lavrenko and
Nigel Goddard
TA:
Boris Mitrovic
The goal of this course is to introduce students to basic algorithms
for learning from examples, focusing on classification and clustering
problems.
This is a level 9 course intended for MSc students and 3rd year
undergraduates.
For an overview of the planned course topics, see the
Course Catalog entry and
the 2011 introductory handout.
Lectures
14:10-15:00 Mondays and Thursdays in
Appleton
Tower, Lecture Theatre 1
Resources
Discussion Forum
Please
sign up for the discussion forum.
The forum will contain detailed lecture notes annotated with questions
and answers. It is monitored by both lectures and the TA, so
if you ask questions here, you are likely to get a much faster
response than if you email the lecturers individually. (However, if
you have issues that should be kept confidential, then of course
please do email the course lecturers.)
Assignments
There will be four equally-weighted assignments together worth 25% of
the mark for the course. These will be released, and submissions due,
according to the schedule below.
Assignment |
Released |
Due |
1 |
Week 3, Monday, 4 p.m. |
Week 4, Friday, 4 p.m. |
2 |
Week 5, Thursday, 4 p.m. |
Week 7, Wednesday, 4 p.m. |
3 |
Week 7, Wednesday, 4 p.m. |
Week 9, Monday, 4 p.m. |
4 |
Week 9, Monday, 4 p.m. |
Week 10, Friday, 4 p.m. |
You are required to submit both an electronic copy and a manual copy
to the ITO by the deadlines. The deadlines are strictly enforced.
Assignment #1
The question sheet is here: a112_1.pdf
[discuss].
It is due at 1600 on Friday in Week 4 (Friday 10 Oct
2014)
by manual submission
to ITO and electronic submission (please see question sheet for detailed
submission instructions). The data files are here:
train_20news_partA.arff,
train_20news_partB.arff
Marks will be returned within two weeks of
the due date.
Assignment #2
The question sheet is here: a112_2.pdf
[discuss].
It is due at 1600 on Wednesday in Week 7 (Wed 29 Oct
2014
by manual submission
to ITO and electronic submission (please see question sheet for detailed
submission instructions). The data files are here:
train_auto_partA.arff,
train_auto_partA_base.arff,
train_auto_partB_numeric.arff,
train_auto_partB_full.arff
Marks will be returned by 1600 on Fri 14 Nov 2014.
Please read the Informatics policy on late submissions and plagiarism.
Tutorials
Tutorials will be in weeks 3, 5, 7 and 9.
Labs
Labs will be in weeks 3,5,7 and 9.
[
Groups, times and rooms]
Week-by-Week listing
(This list is subject to change.)
Week 1
Lectures:
Introduction
[slides].
If you can't access the slides, please
sign up.
Tutorials No tutorials in week 1 (nor week 2)
Readings: Textbook chapters 1, 2
Mathematical preliminaries These
Supplementary Mathematics notes
are from the old Learning from Data course. They are
more difficult than what we will need for IAML, but if
you are happy with them you should have no problem with
the IAML maths level.
Week 2
Lectures:
Thinking about Data
[slides]
[pdf],
Mathematics
[slides],
Basic Probability and Estimation
[slides].
Tutorials: No tutorials in week 2
Readings: Textbook chapters 7.1, 7.2
Week 3
Lectures:
Naive Bayes Classification
[slides]
[pdf]
Tutorial 1: Naive Bayes and data representation [
solutions]
Lab 1: Naive Bayes classification
Readings: Textbook chapters 4.2
Week 4
Lectures:
Decision trees
[slides]
[pdf],
Overfitting and generalization
[slides 1-10]
[pdf]
Tutorials: No labs or tutorials in week 4
Readings: Textbook chapters 3.2, 3.3, 4.3, 6.1, 6.5
Week 5
Lectures:
Linear regression
[slides]
[pdf]
Evaluation
[slides 11-29]
[pdf]
Tutorial 2: Decision Tree and Gaussian Naive Bayes [
solutions]
Lab 2: Attribute selection and Regression
Readings: Textbook chapter 4.6 (but pairwise classification, perceptron learning, Winnow are not required)
Week 6
Lectures:
Logistic Regresssion
[slides]
[pdf],
Optimisation
[slides]
[pdf],
Regularization
[slides],
[discuss],
Support vector machines part I
[slides],
[discuss]
Readings: Textbook chapter 4.6 (but pairwise classification, perceptron learning, Winnow are not required); 6.3 (max margin hyperplane, nonlinear class boundaries),
SVM handout. SV regression is not examinable.
Week 7
Lectures:
Support vector machines Part II
Tutorials: Logistic regression
This page is maintained by
Victor Lavrenko
and Nigel Goddard.