Artificial Intelligence Institute, University of Groningen, The Netherlands



Kunstmatige Intelligentie 2 (KI2)

Artificial Intelligence 2

Academic year: 2008 - 2009




Overview:
This course is intended to introduce topics not handled in the introductory course (KI1) and other advanced topics in AI.

Course status:

Compulsory course in the 3rd study year, period 2, semester 1, 5 ECTS, course code KIB.KI203.

Instructors:
Student Assistants:
Study material:

Lectures:

The lectures will take place on Mondays, 9.00 - 11.00 hrs, in room 267 of Bernoulliborg.

No.
Date Time & Place Title Lecturer
Slides
1
10 nov 2008 9.00 - 11.00
Bernoulliborg 0267
Structural pattern recognition Marius Bulacu
ppt
2
17 nov 2008 9.00 - 11.00
Bernoulliborg 0267
Probabilities revisited
Bayesian learning
Marius Bulacu
Marius Bulacu
ppt
ppt
3
24 nov 2008 9.00 - 11.00
Bernoulliborg 0267
Belief Networks
Computational Learning Theory
Marius Bulacu
Marius Bulacu
ppt
ppt
4
1 dec 2008 9.00 - 11.00
Bernoulliborg 0267
Data Mining
Clustering Algorithms
Marius Bulacu
Marius Bulacu
ppt
ppt
5
8 dec 2008 9.00 - 11.00
Bernoulliborg 0267
Heterogeneous information integration
Hidden Markov Models
Marius Bulacu
Marius Bulacu
ppt
pdf
6
15 dec 2008 9.00 - 11.00
Bernoulliborg 0267
Markov Decision Processes
Reinforcement Learning
Lambert Schomaker
Matthijs Platje
ppt
ppt
7
5 jan 2009 9.00 - 11.00
Bernoulliborg 0267
Partially observable MDP
Overview / conclusions of the KI2 course
Lambert Schomaker
Marius Bulacu
ppt1    ppt2

Assignments:

No.
Title
Observations / Assistant
Deadlines
Instructions
Software
Data
Literature
1
Structural pattern recognition
pen-and-paper assignment, 6 pts
(Theije Visser)
21 nov 2008 (11.00 hrs)
24 nov 2008 (11.00 hrs)
pdf
--
--
--
2
Spam filter using a naive
Bayes text classifier
programming assignment, 12 pts
(Theije Visser)
5 dec 2008 (11.00 hrs)
8 dec 2008 (11.00 hrs)
pdf
java
zip
pdf
3
Clustering of web users: k-means,
Kohonen SOM, leader-follower
programming assignment, 12 pts
(Matthijs Platje)
9 jan 2009 (11.00 hrs)
12 jan 2009 (11.00 hrs)
pdf
java (zip)
zip
pdf

Policy on the assignments:
The first assignment is individual. For the programming assignments, you may form groups of 2 students and work together on the same program. However, the two questions at the end of the programming assignments must be answered separately and individually by each student. Every assignment has two deadlines. All assignments should be completed before the first deadline. Assignments completed in between the two deadlines will only get half of the corresponding number of points. Submissions after the second deadline will not be considered.

Submitting the assignments:
The assignments can be submitted directly on paper to the student assistant in charge. They can also be submitted by e-mail at the following address: ki2 at ai.rug.nl.
Please specify clearly the student name and number in your submissions.

Practical sessions:

The practical sessions will take place on Fridays, 13.00 - 15.00 hrs, in room 228 of Bernoulliborg.

No.
Date Time & Place Title
Assistant
1
14 nov 2008 13.00 - 15.00
5161.0228
Spam filter using a naive Bayes text classifier: introduction Theije Visser
2
21 nov 2008 13.00 - 15.00
5161.0228
Spam filter using a naive Bayes text classifier: learning stage Theije Visser
3
28 nov 2008 13.00 - 15.00
5161.0228
Spam filter using a naive Bayes text classifier: classification stage Theije Visser
4
5 dec 2008 13.00 - 15.00
5161.0228
Clustering of web users: k-means Matthijs Platje
5
12 dec 2008 13.00 - 15.00
5161.0228
Clustering of web users: Kohonen SOM Matthijs Platje
6
19 dec 2008 13.00 - 15.00
5161.0228
Clustering of web users: leader-follower Matthijs Platje

Evaluation of the programming assignments:
The programs will be evaluated on: correct functioning of the executables, style and efficiency of the code, clarity of the comments. All students must separately and individually also answer the two questions regarding the general underlying principles of the assignment.

Literature for the exam:

All the lecture slides are part of the compulsory literature for the exam. In addition, you must study also the materials mentioned below in the table. Note that all AIMA chapters and page numbers are from to the second edition of the book (green cover, 2002).

No.
Title Book / paper, chapter, pages
1
Structural pattern recognition Pattern recognition philosophy paper (by R. Duin and E. Pekalska) - complete
2
Probabilities revisited
Bayesian learning
AIMA ch. 13 'Uncertainty' pp. 462 - 491
--
3
Belief Networks

Computational Learning Theory
AIMA ch. 14 'Probabilistic reasoning' sections 14.1 & 14.2 pp. 492 - 499
AIMA ch. 14 'Probabilistic reasoning' section 14.4 pp. 504 - 506 up to 'The variable elimination algorithm'
AIMA ch. 18 'Learning from observations' sections 18.1 & 18.2 pp. 649 - 653
AIMA ch. 18 'Learning from observations' sections 18.5 pp. 668 - 670 up to 'Learning decision lists'
Kernel methods paper (by K. Muller et al) - pp. 181 - 185 up to 'Supervised learning'
4
Data Mining
Clustering Algorithms
KDD paper (by U. Fayyad, G. Piatetsky-Shapiro and P. Smyth ) - complete
Clustering paper (by A. Jain, M. Murty, P. Flynn) - complete
5
Heterogeneous information integration
Hidden Markov Models
--
AIMA ch. 15 'Probabilistic reasoning over time' sections 15.1, 15.2 & 15.3 pp. 537 - 551
HMM paper (by L. Rabiner)
- pp. 257 - 266 up to 'Types of HMMs'
6
Markov Decision Processes
Reinforcement Learning
AIMA ch. 17 'Making complex decisions' sections 17.1, 17.2 & 17.3 pp. 613 - 625
AIMA ch. 21 'Reinforcement learning' pp. 763 - 789
7
Partially observable MDP
AIMA ch. 17 'Making complex decisions' sections 17.4 pp. 625 - 628

Additional resources:

Web links:

Grading:

Successful completion of all the assignments will give you 30 points. The final exam is worth 70 points. A minimum of 35 points must be obtained in the written exam in order to pass. The final grade will be the total number of points accumulated in the written exam and the assignments divided by 10 and rounded to the nearest integer or half-integer.

Exam schedule:

In order to be able to take the exam, you must register. Participation in the lectures does not register you also automatically for the exam. Bring a calculator with you to the exam.
No.
Date
Time & Place
1
22 jan 2009
14.00 - 17.00
Examenhal, Zernikelaan 7
2
31 mar 2009
(herkansing)
9.00 - 12.00
Examenhal, Zernikelaan 7

Grades:

Available from the secretariat.

KI2 vs Machine Learning:

Given the importance of probabilistic modeling and learning in modern AI, there exists some overlap between KI2 and some topics treated in the Machine Learning (ML) course (optional masters course, code LIX004M05). The areas of overlap are two. First, Bayesian learning is a fundamental topic covered more in depth at KI2 (with programming assignment) and also theoretically treated at ML due to its importance. Second, Hidden Markov Models (HMMs) are introduced at KI2 in the general framework of Markov processes, however they are treaded more in depth at ML with accent on the Expectation - Maximization (EM) training algorithm and with application examples in linguistics. KI2 is a good preparatory course for ML (not a prerequisite). A student taking both courses will have the advantage of a better understanding of these essential topics as they are presented from two different perspectives.


Author: Marius Bulacu
Last update: May 18, 2009