CS 289, Spring 1999
Knowledge Representation and Reasoning
Instructor Stuart Russell
727 Soda Hall,
russell@cs.berkeley.edu,
(510) 642 4964
Office hours Thursday 2.00-5.00.
Lecture: TuTh 12.30-2.00
Location: 320 Soda
Units: 3.
Suggested prerequisites: CS188 or equivalent, or permission of
instructor.
Description
This class will look at formal representations of knowledge
and at reasoning methods that use them. The first half of the course
covers logical methods for inference and decision making,
while the second half covers probabilistic methods.
Topics will include
- Representation
and reasoning in propositional logic, including efficient model-finding algorithms.
- Representation and reasoning
in first-order logic, including logic programming and resolution.
- Reasoning about action, time, and knowledge.
- Logical planning systems; applications to spacecraft and softbots.
- Probabilistic representation and reasoning
in Bayesian networks.
- Probabilistic temporal reasoning, including hidden Markov models, Kalman filters, dynamic Bayesian networks, and object identification.
- Combining logic and probability.
- Decision theory, including multiattribute utility theory.
- Probabilistic planning in Markov decision processes, including
approximate methods (reinforcement learning) and hierarchical methods.
- Information value, control of reasoning, and bounded rationality.
In most cases we will be concerned with complexity and completeness
as well as implementations. Applications to be studied
range from PC diagnosis and freeway traffic surveillance
to robotic tour guides and autonomous spacecraft.
What will actually happen
The class will meet twice a week; discussion will focus on the
readings given in the accompanying reading
list. There will be three assignments (20% each) combining written
work with simple implementations, and a term project (40%)
consisting of a substantial project or analytical paper.
Books: Russell and Norvig, Artificial
Intelligence: A Modern Approach, Prentice Hall, 1995.
Course reader, available soon.
Handouts