CSC581 Homepage

Topics in AI: Introduction to Machine Learning with Support Vector Machines

Spring 2007

Description:

Support vector machines (SVMs) belong to a new class of machine learning algorithms with their origins firmly rooted in statistical learning theory. Due to the strong theoretical foundation these algorithms possess desirable properties such as the ability to learn from very small sample sets and a firm estimation of the generalization capacity of the learned model. These properties make this new class of learning algorithms extremely attractive to the practitioner who is frequently faced with "not enough data" and needs to understand "how good a constructed model" actually is. The fact that SVMs have surpassed the performance of artificial neural networks in many areas such as text categorization and speech recognition bears witness to the power of this new class of learning algorithms.

This course is an introduction to machine learning and SVMs. We begin by framing the notion of machine learning and then develop basic concepts such as hyperplanes, features spaces and kernels necessary for the construction of SVMs. Once the theoretical groundwork has been laid we look at practical examples where this class of algorithms can been applied. When applied to tabulated data, machine learning can be viewed as an area of computational statistics (as opposed to learning as part of AI where the learning algorithm is part of a larger problem solving system). In this course we take the computational statistics view and apply SVMs to tabulated (real world) data. We will use the statistical computing environment R for our experiments.

This is an exploratory, graduate level course, therefore strong class participation is expected. Readings will be mainly assigned from a book draft, chapters will be made available online as we move forward. Assignments will consist of problem sets and programming assignments using the open source R environment. For the midterm exam you are expected to write implement a simple SVM algorithm and demonstrate that it works. For the final examination you are expected to build an SVM model on a given data set and write up a report analyzing your findings.

Final Projects due on May 10th @ 2pm in my office (see problem sheet below)



NOTE: Chapter 10 is coming...perhaps not...I have some rough notes, but I think the lecture notes are much more comprehensive at this point...

Announcements:

[4/23/07] Posted the final problem sheet.
[4/23/07] Posted the paper on nu-SVMs.
[4/11/07] For those folks who are taking the CSC581 course for 4 credits I just posted an additional assignment. This is due on 4/30 in my office.
[4/11/07] Posted assignment #8
[3/31/07] Hint for the homework: before splitting into k folds for cross-validation you should randomize your dataset in case it is sorted by some criterion. See the midterm for the R code.
[4/3/07] Posted Ron Kohavi's paper on cross-validation.
[3/29/07] Posted chap 9 and Platt's paper.
[3/29/07] Posted assignment #7
[3/12/07] posted the midterm
[3/9/07] posted chapter 8
[3/7/07] posted assignment #6
[3/7/07] posted the kernel-adatron paper.
[3/4/07] Posted chapter 7 on support vector machines...apologies for the delay.
[2/28/07] NOTE: **no** homework due for next week (except for the reading, chapter to follow), I decided against the dual perceptron.
[2/23/07] posted chapter 6
[2/22/07] posted assignment #5
[2/14/07] posted chapter 5
[2/12/07] Posted assignemnt #4
[2/8/07] Posted chapter 4
[2/7/07] Posted chapter 3
[2/5/07] Posted assignment #3
[2/5/07] Schedule change: class will start at 5:30pm beginning 2/6.
[1/25/07] Room change: starting immediately we will be in Washburn Rm 220
[1/22/07] Welcome!

Documents of Interest:

Data Sets:

Many of the packages above have accompanying data sets.  But the premier source for experimental machine learning data sets is the UCI  Machine Learning Repository. The Statlib library at CMU is another great place to look for data.

Assignments:

Instructor:

Dr. Lutz Hamel
Tyler, Rm 251
Office Hours:  TBA
email: lutz at inductive dash reasoning dot com