COMP30411: Knowledge Representation

This is Bijan Parisa's page for the Fall 2007 instance of COMP30411. Links to slides for the classes appear on the schedule, though I include links to my own slides (with a bit of commentary) here as well. 

Lectures

1. Exam Prep
2. Logic Engineering (4) (20-11-2007)
3. Logic Engineering (3) (15-11-2007)
4. Logic Engineering (2) (13-11-2007)
5. Logic Engineering (1) (08-11-2007)
6. Some Commonsense (25-10-2007)
7. What is a Knowledge Representation? (11-10-2007)
• As promised, you can download the PDF version for printing.
• This lecture explicates the classic paper, "What is a Knowledge Representation?" by Randall Davis, Howard Shrobe, and Peter Szolovits.
• They less define a KR than characterize it. This is somewhat amusing given that we're all about definitions (e.g., necessary and sufficient conditions).
• They characterize it in terms of 5 roles it can play. See the slide for details.
• I think two things they got exactly right is that a KR is a medium for effective computation and for human expression. This is (mostly) true no matter what your focus, though the commonsense representation tradition (see, in particular, the Second Naive Physics Manifesto by Patrick Hayes) runs somewhat counter to the first (prefering strongly, I would say, the human expression aspect).
8. More defining (09-10-2007)
9. Moving toward formalization (04-10-2007)
• We didn't quite get through the lecture because of a rather extended discussion of CW1, in particular, the difference between a term which is a "Modifer" and one which is a "Self-Standing"
• The main point there, I think, is two fold:
• These are organizational rather than instrinsic. That is, to some extent these are features of how we choose to build our model.
• Modifers tend to be somewhat adjectival, or tend to form compound nouns.
• "Female Cat" comes to mind.
10. Intro to KR (25-09-2007)
• This lecture is notable, I think, for the regexp analogy. I took the basic technique (at least, the comparison of knowledge oriented code vs. non-knowledge oriented code) from a book, The Logic of Knowledge Bases. The basic point is easy to grasp: Having a nice representation for what you are trying to express can help you write clearer, cleaner, more maintainable, and possibly better running programs. Having an optimizable query language is likely to result in generally better performance (across all queries and all query writers) that what results from forcing everyone to write custom code for each query. Perhaps more importantly, improving the query engine will improve all the queries, whereas if everyone is writing custom code that code there's no clear point to improve. Let's not even get into correctness!
• Much of this is fairly generic. I make this point too by bringing in the "information systems" perspective. But some of this is also the argument people make for any declarative language. The notions of "declarative" and (that is, vs.) "imperative" (e.g., in programming languages) is not the most useful notion esp. when they play proxy for "good" vs. "bad" (specifically, "declarative" has become such a praise term that its substantive content is suffering; part of this is the historical accident that many "declarative" languages were merely more declarative than others, or were declarative up to a certain point).
• It's possible that KR as a specific subfield will mostly disappear in another 10 years. In my lecture I speak of three (non-exhaustive) foci of KR, commonsense representation, scientific representation, and conceptual modeling. Since KR is generally an add-on to conceptual modeling (or rather, KR advocates like to validate their work by appeal to the applicability to conceptual modeling), it's easy to see how that use might merge into CM in general. Commonsense representation seems pretty generally moribund, at least as a grand project. Perhaps "in a quiet phase" would be better wording. Scientific representation is a burgeoning field and it might just stay self-standing. But KR in the service of bio-informatics (e.g., via ontologies) alone will have a very different flavor than the KR we know now.