INLS-110-40 – Machine Learning
 Spring 2002, MW 8-9:15, Room 208
UNC-CH SILS
Bob Losee, Manning 302
962-7150, losee@ils.unc.edu
 

Brief Description

Learning algorithms that can be implemented on computers.  Emphasis will be on statistical methods for both supervised and unsupervised learning, pattern recognition, and clustering.  Non-statistical methods will be discussed and compared to statistical methods

Prerequisites

INLS 150 and 172 are recommended although not required.  While there is no mathematical prerequisite, students should anticipate reading materials describing a variety of mathematical techniques consistent with a variety of mathematical paradigms.  Math phobic students will be uncomfortable in this course; students are not expected to understand everything or even most that they read, but are expected to appreciate the different approaches.

Textbook

Hastie, Trevor, Tibshirani, Robert, and Friedman, Jerome.  The Elements of Statistical Learning, Springer, 2001. (Denoted below as HTF) 

 

Students will find it beneficial to obtain a flexible programming package (Mathematica or S) designed for statistical methods.  Some Mathematica (produced by WRI) will be used in class and the book web site has some code in S. Mathematica is available for purchasing in the RAM shop, and is installed on PCs in our lab, as well as ruby, our unix workstation.  The author will present some procedures in class using Mathematica.  The author does not anticipate presenting anything in S (although it is a fine package)

Outline and Readings (Readings preceded by * are optional)

Topic

Readings

Introduction and Probability Basics

HTF Ch. 1

Supervised Learning

HTF Ch. 2

Linear Methods for Regression

HTF Ch. 3

Linear Methods for Classification

HTF Ch. 4

Bayesian Methods

Bishop Ch. 1, 2.3

Theodoridis Ch. 2.1-2.4

Specific Distributions

Pratt et al. Ch 9, 11.

Basis Expansion and Regularization

HTF Ch. 5

Kernel Methods

HTF Ch. 6

Model Assessment and Selection

HTF Ch. 7

Neural Networks

HTF Ch. 11

Bishop Ch. 3

Support Vector Machines

HTF Ch. 12

Unsupervised Learning

HTF Ch. 14

Evolutionary and Genetic Algorithms

 

Sequences and HMMs

Baldi and Brunak,
Ch. 7 & 8

Learning Game Strategies

Poundstone, Ch. 1, 12, 13.

 

Semester Projects

Students are expected to conduct four small studies using machine learning techniques.  Students will develop one or more databases to be used for these projects.

 

The first project should implement a regression (without using the “regression” command found in the programming package the student is using.)

 

The second project should use neural networks

 

The third project should use unsupervised learning

 

The fourth project should use genetic algorithms, learning sequences, or game strategies

Evaluation

Class participation & homework 60%,
Projects 40%

Bibliography

Baldi, Pierre, and Brunak, Soren.  Bioinformatics, Second Edition.  MIT Press, 2001.

 

Bishop, Christopher.  Neural Networks for Pattern Recognition.  Oxford U. Press, 1995.

 

(HTF) Hastie, Trevor, Tibshirani, Robert, and Friedman, Jerome.  The Elements of Statistical Learning, Springer, 2001.

 

Poundstone, William.  Prisoner’s Dilemma, Doubleday, 1992.

 

Pratt, John, Raiffa, Howard, and Schlaifer, Robert.  Introduction to Statistical Decision Theory.  MIT Press, 1995.

 

Theodoridis, Sergios and Koutroumbas, Konstantinos.  Pattern Recognition.  Academic Press, 1999.

 

Webb, Andrew.  Statistical Pattern Recognition, Arnold Press and Oxford U. Press, 1999.

Honor Code

Students should familiarize themselves with the University of North Carolina at Chapel Hill Honor Code, which is described in University publications.  It should be noted that in this course, students may receive (and provide) some assistance with general problem solving techniques and the readings.  Students should NOT receive (or provide) major creative assistance on the class project.

 

 

Last updated:  February 20, 2002