CPSC 422 Intelligent Systems

Winter Session 2004/2005 Term 2

Course Description
People & Consulting Hours
Grading Scheme
Lecture Overview (see slides used in class).
Assignments
Lab and Tutorial Schedule
Textbook and Suggested References
Resources

Course Description

Building on material from CS 312 and CS 322, this course explores the science and technology developed for designing and implementing intelligent systems. Essentially, CS 322 developed the logical approach to agent design in the Good Old Fashioned Artificial Intelligence and Robotics (GOFAIR) model. In GOFAIR, we assumed that there is a single agent operating in a completely (pre-)known, deterministic environment where there are no other actors. We move beyond that by systematically lifting those restrictions so we need to deal with vague, incomplete, uncertain, possibly incorrect, beliefs about the world where there may be several other agents. The course will take an agent-centered approach to these issues. This course will assume that the students are proficient in logic programming (e.g., as covered in CS312).

The following topics will be addressed:

Robot Control

Planning Under Uncertainty

Reinforcement Learning

Assumption-based Reasoning: abduction & default reasoning; diagnosis & design;

Using Uncertain Knowledge: probability, Bayesian belief networks, inference, learning belief networks, dynamic belief networks, perception, hidden Markov models

Decision making; utility, decision networks,

Multi-agent systems

Ontologies, Semantic web, OWL

People & Consulting Hours

Instructor:

David Poole
Email: poole@cs.ubc.ca
Office: CICSR 127
Office Hours: Tuesdays 12:30-1:50 or by appointment.

Teaching Assistants:

Michael Chiang mchc@cs.ubc.ca Office hours: Thursdays 12:50-1:50 in CS Atrium or by appointment
Frank Hutter hutter@cs.ubc.ca Office Hours: Wednesdays, 9:50-10:50 Room CICSR/CS 341 or by appointment

Grading Scheme

The following is a rough guideline only. The final grading scheme may vary slightly.

Written Assignments 20%

Midterm Exam 20%

Group Project 20%

Final Exam 40%

See the course outline or (pdf) for more details, particularly about plagiarism.

Lecture Overview / Lecture Notes

The lectures will follow the material of Chapters 6, 7, 9, 10, 11 and 12 of the textbook. Additional material will be supplied (particularly on planning under uncertainty and reinforcement learning). You can also get copies of the slides used in class,

Assignments

There will be about 5 assignments. The current plan is for the assignments to cover the following topics:

Assignment 1 (pdf) Robot control. Due to possible bugs in the applet there is an optional alternate question 2(a). We now have a solution.

Assignment 2 (pdf) Decision-theoretic planning. We also have a solution and corresponding Java code (thanks to David, Ian and Yavar). Note that you didn't need such a slick solution to get full marks.

Assignment 3 (pdf) Reinforcement Learning

Assignment 4 (pdf) Belief Networks

Assignment 5 (pdf) Decision Networks. Here is the CIspace representation of question 2.

and one literature review and a group project.

Lab and Tutorial Schedule

There are no labs or tutorials scheduled for this course. Students should make use of Computer Science computing facilities or any PC running Mac OS, Linux or Windows to complete homework assignments. (You will only need Prolog, (e.g., SWI prolog) and Java runtime).

Textbook and Suggested References

Required Textbook:

Computational Intelligence: A Logical Approach. Poole, Mack worth and Goebel. Oxford University Press, 1998. Available at the bookstore. (If you want to but it cheaper, there is a link at the bottom of the book page for price comparisons).

Suggested General Reference:

Artificial Intelligence: A Modern Approach. 2nd edition, Russell and Norvig, 2003. Available at the Reading Room.

Resources

Email
David if you have questions or comments.
The WebCT site for this course is here. Use it for discussion and grades.

There is a newsgroup for general stuff like buying/selling the text (but use Web CT for discussion if you are in the course) news:ubc.courses.cpsc.422

The textbook's web site contains valuable pointers such as all of the code from the book, solved exam-type questions and slides from lectures.

Check out CIspace: tools for learning Computational Intelligence. We have applets for learning about graph searching, constraint satisfaction problem solving, stochastic local search, neural network learning, robot control, belief and decision networks and definite clause deduction.

CPSC 422 (Winter 2004/2005, Term 2)