Artificial Intelligence Seminar


Winter 1996

Wednesdays 14:00 - 15:20, Deschutes 200


Tractable Approaches to Logical Deduction


Since logical deduction is NP-complete in the propositional case and undecidable in general, building practical reasoning systems presents difficult choices. In some cases, sound and complete methods may be refined to the point where they are adequate for a particular class of problems (this was a topic of James Crawford's AI seminar last winter). In other cases, however, something must be given up in order to achieve reasonable performance. The obvious candidates are expressive power, soundness, and completeness. Each of these has been the topic of extensive investigation in the quest to build practical reasoners.

We will cover a variety of approaches to tractable knowledge representation and reasoning, ranging through languages with restricted expressive power, "knowledge compilation", multi-valued reasoning and other incomplete techniques, and approximation.

We will likely begin with Lipski's early work on approximate answers to database queries, which foreshadows some of the key ideas that have recently attracted interest, consider logics of relevance and explicit belief from the 1980's, and end with compilation, approximation, and hierarchical schemes.


The class schedule and reading list are now available.
Back to David Etherington's home page.