COSC 4555/5555 MACHINE LEARNING

Spring 2006

TR 11:00-12:15, Engineering Hall, Room 4059

Instructor: Prof. D. Spears	Phone/Office: 766-5485 / 4087 ENG
Office Hours: TBD	Email:dspears arobase cs.uwyo.edu

Textbook: Machine Learning

Author: Tom M. Mitchell

Publisher: McGraw-Hill, 1997

ISBN: 0-07-042807-7

Course Description:

The goal of machine learning is to program machines to learn and improve their performance on their own, based on experience and/or data. Learning should be semi-autonomous, i.e., it should require little or no human intervention while the program is executing. The first part of this course will cover the main machine learning algorithms. To reinforce the students' understanding of these algorithms, there will be homework assignments and a midterm exam. The second part of the course will begin with lectures on how to evaluate learning, and the theory associated with machine learning. It will then involve discussions about machine learning applications, including self-improving robots, scientific discovery by machines, and evolving artificial brains. The emphasis during the second part of the course will be on class discussions and individual student projects.

Prerequisites:

The prerequisites are: (1) experience writing software programs, and (2) basic probability and calculus (knowing how to compute derivatives). COSC 4550/5550 (Artificial Intelligence) is recommended, but not required

Tentative Schedule:

Part of the course will be lectures, and part will be discussions about papers. Students will take turns leading discussions on the papers. Relevant chapters of the textbook to read for each topic are in parentheses after the topic title in the syllabus below.

*** COURSE PREREQUISITE: ***

The basics of probability

GUEST LECTURES:
- Guest lecture by Jinwook Shin on "Using induction to learn the basic building blocks of cyber attacks" on Tuesday, January 17.
- Guest lecture by Suranga Hettiarachchi on "Genetic algorithms and advice on how to apply them to problems" on Thursday, January 19.
*** PART I OF THE COURSE: The basic learning algorithms ***

1. INTRODUCTION
Introduction to Machine Learning (Chapter 1)
Background on search

2. INDUCTIVE, SUPERVISED CONCEPT LEARNING (CLASSIFICATION)
Overview of inductive inference (Chapter 2)
Version Spaces and the Candidate Elimination Algorithm (Chapter 2)
Decision trees (Chapter 3)

3. LEARNING WITH NEURAL NETWORKS
Artificial neural networks (ANNs) and backpropagation to learn their weights (Chapter 4)

4. INSTANCE-BASED LEARNING AND CASE-BASED REASONING (CBR)
Instance-based learning (IBL) (Chapter 8)
Presentation about an application of CBR to Intrusion Detection

5. ANALYTICAL "EXPLANATION-BASED" LEARNING
Explanation-based learning (EBL) using deductive inference and a domain theory (Chapter 11)
Discussion on a paper about an application of EBL

6. LEARNING BY REINFORCING BEHAVIOR
Reinforcement learning (RL) (Chapter 13)

7. BAYESIAN LEARNING
Bayesian learning (Chapter 6)

8. UNSUPERVISED LEARNING (CLUSTERING)
Clustering
Discussion on papers about clustering

MIDTERM EXAM ON THE LEARNING ALGORITHMS

*** PART II OF THE COURSE: Evaluation, theory, and applications ***

DISCUSSION ON SEMI-TERM PROJECTS, AND PROJECT PROPOSALS

9. EVOLUTIONARY LEARNING (Optional)
Evolutionary algorithms (Chapter 9)

10. EVALUATING LEARNING
Evaluating and comparing learning algorithms (Chapter 5)
(Optional) Statistical significance tests

11. MACHINE LEARNING THEORY
Computational learning theory (Chapter 7)
Is learning possible? The No-Free-Lunch Theorem

12. APPLICATIONS OF MACHINE LEARNING

The scope of this portion of the course will depend on the time remaining, as well as the interests of the students.
Using neural networks to improve reinforcement learners (paper)
Cooperative multi-agent reinforcement learning (slides on Utility; papers)
Learning from massive amounts of data: Data mining (papers)
Bioinformatics, biomedicine, and the role of machine learning (papers)
Computational scientific discovery (papers)
Cataloguing the stars and galaxies (papers)
Evolving artificial brains (papers)
Self-improving robot soccer teams (papers)
A cognitive model of human learning (papers)
"Safe" learning (papers)

Homework:

There will be written/programming homework assignments during the first part of the course.

Exams:

Midterm exam, date TBD, during class. Will cover the basics of machine learning from the lectures and the homeworks. There will be no final exam, only a midterm.

Semi-Term Projects:

During the latter half of the course, students will work independently on a project of their choice. The topic can be anything that involves *any* type of learning (including evolutionary algorithms learning, for those who have studied this in the AI or EA course). For students who are working on a thesis that involves learning, the project may be on your thesis topic. During finals week, we will have two afternoons devoted to a "conference" for students to give final project presentations. The agenda for the conference will consist of 20-minute presentations by each student on their project, along with questions and discussion. The presentation format will be a talk with slides, followed by a demo. Fancy graphics is not required -- simple ascii graphics is acceptable. The grade will be based on the machine learning content and creativity, not the graphics sophistication. A final written report on the project is also due on the day of the conference. The level of sophistication of term projects is expected to be higher for graduate than undergraduate students.

Grading:

The various required work in this class will be counted towards your final grade as follows:

Homework - 25%
Midterm Exam - 25%
Class Participation - 20%
Semi-Term Project - 30%

Grading will be on a curve. The professor reserves the right to alter the grading scheme or to take extenuating circumstances into account when assigning grades. Discussion of the course material among students is encouraged, although students are expected to write up their own assignments and programs, unless it is a team project. Copying code, homework solutions, or exam material from any source at all (unless explicitly permitted by the instructor) will most likely result in failing the course. Academic dishonesty will be treated in accordance with the strictest university standards. Students are urged to read University regulation 803, section 3 defines academic dishonesty.