Machine Learning

Course Number: LIX004M5
Instructor: Jörg Tiedemann, Alfa-informatica, tiedeman@let.rug.nl

Announcements Autumn 2006

Re-examination results
All results available now!
Results of the 3rd lab. Just to remind you: you need an average of 6 for all labs together to pass the course. Remember that the last lab counts double! And, of course, you can get the corrected labs back if you like. I forgot that already last time in class where I had the print-outs with me. I will take them with me on monday again but you can also pick them up from my office if you like. (But don't expect very detailed comments on each of them - unfortunately, I don't have the time for that - sorry)
Here is an example exam (skip the SVM parts)
about example 1 on the crazy soft drink machine in chapter 9 of Manning & Schutze
Modified HMM package with the -i flag (deadine for lab 4 moved to 23/10 2006)
Results of lab 2 are on-line
You may work in groups in Lab 3 and 4 (see the instructions)
Results of the first lab are out
Sorry for the confusion about conditionally independence in Bayes nets in class. Look at this for some further explanations.
It's still possible to move the day for the exam. Is it still the best for everybody to leave it on the same day as scheduled below? Let me know otherwise.
Sign up for the exam! You have to do that 2 weeks in advance! (week 5 or 6)

Schedule

Sept 4 - Oct. 20
Lectures: Academiegebouw A12, Mon. 13:15-15:00 pm
Labs: Fri. 9:15-11:00 am in Harmonie.12.102C. Four labs, starting Fri. Sept. 15 (2nd week, not every week!) (Students' Unix Room).
Exam: Friday, 27 Oct. 09:00 - 12:00 in the Academiegebouw AZERN

Prerequisites:

The course is open to students in Computer Science, Artificial Intelligence and Information Science. Required background consists of programming ability, elementary statistics, status as 2nd year student (or higher) in study program.

Coursework, Grades, Responsibilities

There are four required labs, which count as 50% of the grade, 10% each except the last one (20%). A written exam counts for the remaining 50%. Students must accept the responsibilities outlined in the general statement of intellectual responsibility for Information Science students.

Book

Tom Mitchell Machine Learning New York:McGraw-Hill, 1997.
Overhead Sheets from the book: www.cs.cmu.edu/~tom/mlbook-chapter-slides.html

Exam

The exam is open book. Books, notes and calculators may be used.

Guidelines for writing lab reports

All reports must be your own work. No copying.
You may write your reports in English or Dutch. Be assured that you will not lose marks simply for making mistakes in your English (or Dutch - but I can't judge that anyway ;-))
Please submit reports by email to me (tiedeman@let.rug.nl) in either postscript (.ps), plain text (.txt) or pdf format
or on paper in my mail box in the Harmoniegebouw, 4th floor
Each report is marked out of 10. You have one week to hand in the report except for the last one (2 weeks). If you are up to 1 week late, the report will get half marks. More than 1 week late and the report gets no marks.
Your report should include the questions for the relevant assignment (but not the introduction).
Give concise but complete answers to the questions.

Preliminary program

Organization, Inspiration &
Intro Machine Learning, Mitchell Ch.1, Ch.5
slides

Inductive learning as search with bias, Mitchell Ch.2
Decision Trees, Mitchell Ch.3
slides

Lab 1 - Decision Trees (deadline: 21-09-2006)
Instance-Based Learning, Mitchell, Ch.8
(slides)

Lab 2 - Instance-Based Learning (deadline: 28-09-2006)
Bayesian learning, Mitchell Ch.6
(slides)

Lab 3 - Learner comparison/combination (deadline: 05-10-2006)
Sequential data & Markov Models
Manning & Schütze,Ch. 9,
Bilmes: What HMMs can do
(slides)

Lab 4 - Markov Models (deadline: 23-10-2006 (NEW!))
Maximum Entropy models
Combining Learners
Berger et al: A Maximum Entropy Approach to NLP
Dieterich: Ensemble Methods in Machine Learning
Alpaydin: Techniques for Combining Multiple Learners
(slides)

Genetic Algorithms, Mitchell, Ch. 9
(slides)

About WEKA

Some information about WEKA and how to use it (old):

using WEKA
WEKAs file format
interpreting WEKA output
WEKA homepage (documentation, Wiki, ...)