What is a knowledge representation? Why are issues in knowledge representation important for application builders? What is the relationship between knowledge and data, between knowledge bases and data bases? How do I build an ontology?

These are some of the questions we will pursue in this course. We will read and discuss the primary literature in artificial intelligence and biomedical informatics as they relate to these questions over the past 20 years or so. As per the schedule below, topics covered will include: frame-based systems, description logics, automatic theorem proving, complexity vs. tractability, ontologies, rule-based systems, and a variety of applications in the biomedical domain. Although we will cover a fair amount of computer science (primarily artificial intelligence), the emphasis will be on the implications of these results on the biomedical and health informatics field.

Detailed Reading List & Syllabus

E-reserves listing of papers

Evaluation:

This course will be evaluated on the basis of homework and formal writing assignments, class participation, and on a final project. The writing assignments will be three 2-page "reaction papers" in response to selected readings. (What's a reaction paper?) There will also be one quiz and two "hands-on" assignments where I ask students to carry out tasks with some of the knowledge-based systems or knowledge representational formalisms that we cover in the course. There will be fairly extensive reading assignments: about 2-3 journal-length papers (or equivalent) per week. I expect this class to be a discussion-oriented seminar, so class participation will also count toward grading. Finally, students must present a 15-30 minute oral presentation that summarizes their work on a final project. Details will be posted later.

For Spring '04, the final grades will be based on the following percentages: Final project: 35%, Reaction papers: 30%, Homeworks: 15%, Quiz: 15%, Class participation: 5%

The specification for the final project is now available.

Learning objectives:

1. Have a passing familiarity with basic KR formalisms: (a) FOL, (b) Rules, (c) Semantic nets, and (d) Frames.
2. Understand inference and the implications for tractable KR systems.
3. Have a knowledge-level understanding of some of the well-known knowledge-based applications.
4. Be able to critically read and review primary literature about KR in biomedical informatics. This includes the ability of synthesis: connecting ideas across multiple readings.
5. Be able to write English clearly. (More generally, communicating clearly and effectively should be a life-long objective. In this course, I hope to help make some progress toward this goal.)

Prerequisites: Any course in Artificial Intelligence, or permission of instructor. In order to quickly teach inference and theorem-proving, students should have a prior knowledge of at least the notation of 1st-order predicate calculus and logic.

Academic Honesty - To prevent possible misunderstandings, students must read the University of Washington's Statement on Academic Honesty. All students are expected to follow high standards of academic honesty in all aspects of this course -- this is especially true of your written reaction papers.

Approximate schedule:

My goal is to interleave theory and applications -- to define the concepts from an abstract point of view, and then to see what the practical implications of those concepts might be in the biomedical domain. Another way to look at the schedule below is that we'll sneak in examples of Frame-based systems (Protege), Rule-based systems (Mycin), and Semantic nets (RDF). We'll also present various inferencing capabilities (from description logics to relational algebra). Then, in week 7, we'll be prepared to present and discuss the theoretical tractability vs. expressivity tradeoff across all of the these systems.

Below is approximate, and subject to change.