!! FILLED !!

Post-Doctoral Research Fellowship In Computer Science

University of Alberta
Edmonton, Canada

Applications are invited for a one-year (renewable) fellowship to work in the areas of

The postdoc will be required to carry out high-quality research, over the entire gambit of research activities, ranging from formal theoretic explorations, to concrete implementations and empirical testing. S/he may work with various local companies, including

as well as the design and development of the AI PlayGround. This is in addition to pushing on your own "curiousity-driven" and "technology push" ideas; see also RG's ideas.

Candidates should have a Ph.D. in Computer Science or the equivalent. Previous research excellence and strong productivity in addition to good computing background is essential. If interested, please send

(to arrive ASAP) to:

Russell Greiner
Department of Computing Science
Athabasca Hall 359
University of Alberta
Edmonton, AB T6G 2H1

Email: greiner@cs.ualberta.ca
Phone: (780) 492-5461
Fax: (780) 492-1071

Review of applications will begin immediately, and will continue until the position is filled.

Electronic submissions -- in plain text or PostScript -- are encouraged.

We also encourage applicants to apply for various additional sources of funding, including

and well as other possible sources.

See http://www.cs.ualberta.ca/~greiner/ for more information about my research, and http://www.cs.ualberta.ca/ for more information about the department in general.

Edmonton is also a great place to live! See AboutEdmonton for more information about the city!

Other Comments

While I (Russ Greiner) will be the primary contact, several others at UofAlberta have related interests, including:

Peter van Beek constraint satisfaction, planning
Bill Armstrong adaptive logic networks
Renee Elio cognitive modeling, agent communication
Randy Goebel default reasoning, and other representation issues
Jonathan Schaeffer game playing(Chinook), search, parallel systems
Tony Marsland game playing, search

as well as on-going activities in logic programming, vision, robotics, and many collaborations with others in areas outside of AI (including philosophy, psychology, ...); see also AI Lab HomePage.

We are most interested in a researcher who can do high quality research, which results in publications and perhaps distributable code; see SoftwarePage. I also anticipate getting some money from an industrial company to work on some specific theoretical aspects of certain funded applications. Here, the post-doc and I will be expected to apply the ideas, and code, that we develop to their datasets. (I view this as a wonderful opportunity: getting data, and specific problems, is often one of the hardest aspects of research!) Of course, I will also make sure that the funders expect research from us, rather than development.

Also, the postdoc will have the option of teaching a course, as a way to supplement his/her income.

Challenge Problem

The raison d'être for building a Bayesian Net is to answer queries; this often involves computing P( h | e), the posterior probability of the hypothesis h, conditioned on the observations e. Eg, a patient may want to know the probability that he is suffering from a heart problem, given the set of recent sensor measurements. Unfortunately, it can take an extremely long time to produce answers to queries, both in theory (it is NP-hard) and in practice. Fortunately, there are often ways to make this computation more efficient, in many situations. In particular, even if one query-answering algorithm (QA) is slow for a specified query, another algorithm may be quite efficient for the same query. Moreover, different BNs can express the same distribution. Even if a particular QA algorithm is slow for a given query when using one BN, that same algorithm may be efficient for this query, in a different, but equivalent BN.

Our challenge, then, is to find the "most efficient" BN/QA combination -- i.e., determine

to minimize the expected time to answer queries, over the distribution of queries that will be encountered.

I am very interested in hearing any specific ideas on how to solve this problem.

See these articles for one possible framework for posing this problem.