The Sixth Bayesian Modelling Applications Workshop

Call for contributions

July 9th, 2008
Helsinki, Finland

Special theme: How biased are our numbers ?


OBJECTIVE

As in previous years, the workshop will provide a focused but informal forum for fruitful exchanges among theorists, practitioners and tool developers. Discussions may cover research questions and insights, methodologies, techniques, and experiences with applications of Bayesian models to any particular problem domain. Contributions will be compiled into separate workshop proceedings. In addition, based on the response at the workshop, our intention is to publish a subset of the papers in a special issue of an appropriate journal.

WORKSHOP THEME

This year we are especially encouraging contributions that address our special theme

"HOW BIASED ARE OUR NUMBERS ?"

We encourage papers that focus on issues relating to (probability) biases in applications of Bayesian networks. Examples include, but are not restricted to, the following issues.

In constructing a Bayesian model, probabilistic information is required for establishing the numerical parameters of the model. This probabilistic information can be obtained using available datasets, human experts, a mix of these, or from yet other sources. All these sources of probabilistic information are known to be biased: datasets, for example, are often gathered for a different purpose than for constructing the model and therefore exhibit a selection bias; expert probability judgments are known to be biased as a result of the heuristics humans use for assessing probabilities.

In acquiring the numerical parameters for a Bayesian model, how can the biases in the sources of probabilistic information be identified? How can the degree of bias be established? What is the effect of biases on the resulting model and its behaviour? Is it possible to correct for biases, and if so, how?

In eliciting probabilities from human experts, an additional source of possible problems is the method used for elicitation. Dedicated techniques are being designed for this purpose, but do these techniques forestall, for example, biases and over-commitments of the resulting model? Are these techniques efficient and easy to use? Do they scale up to building large models? Are the resulting assessments well-calibrated? How can you tell? How can knowledge obtained from multiple experts, or multiple sources in general, be combined? What are the benefits and drawbacks of doing so?

In verifying the probability assessments and behaviour of the model under construction, biases in the numbers and in the interpretation of these numbers can be expected. What type of bias can we encounter? How can you tell? What can you do about it? What kind of probabilistic information, possibly computed from the model under construction, do you feed back to, for example, a human expert? How do you communicate such information? Is it interpreted as intended?

GENERAL FOCUS

For the 2008 Bayesian Modelling Applications Workshop we encourage submissions that provide insight in and examples of encountered biases in sources of probabilistic information, or introduced by the methods, new or existing, for obtaining and communicating the numbers. We in addition encourage proposals of methods for identifying biases in the numbers, for establishing their effect on model behaviour, and would appreciate solutions to all posed and related problems.

In addition to submissions addressing the above factors, we also welcome contributions to the overall focus of Bayesian modelling. Submissions addressing novel applications are particularly encouraged.

FORMAT

The workshop will take place on Wednesday, July 9th, the day before the UAI main conference starts. The workshop will consist of sessions devoted to issues in developing and using Bayesian models, in the form of presentations and open discussions. We will conclude with a plenary panel discussion to summarise the issues raised during the day and to consider plans for the next edition.

ARRANGEMENTS

The workshop and the UAI main conference are co-located with the COLT and ICML conferences at the University of Helsinki, Finland. A separate workshop-registration fee is required.

TO CONTRIBUTE

Submissions of 3-8 pages in PDF format, preferably in the same style as for the main conference, are solicited. The contribution should raise questions and offer results that can be presented and discussed at the workshop. Participants are encouraged to apply jointly with members of other disciplines with whom they have collaborated.

Submissions are peer-reviewed and should be sent to UAIworkshop2008@cs.uu.nl with as subject: "Submission [authors' names(s)]", or something similar. Inquiries can de made at the same address.

WORKSHOP COMMITTEE

Suzanne M. Mahoney Innovative Decisions, Inc. Co-chair
Silja Renooij Utrecht University Co-chair
Hermi J.M. Tabachneck-Schijf Utrecht University Co-chair

John Mark Agosta Intel Research
Russell Almond Educational Testing Service
Dennis M. Buede Innovative Decisions, Inc.
Marek Druzdzel University of Pittsburgh
Linda van der Gaag Utrecht University
Judy Goldsmith University of Kentucky
Sean Guarino Charles River Analytics
Kathryn Laskey George Mason University
Ann Nicholson Monash University
Jonathan Pfautz Massachusetts Institute of Technology
Juan-Diego Zapata-Rivera Educational Testing Service

IMPORTANT DATES

Submission of contributions: Monday March 31st, 2008
Notification of selections: Friday April 25th, 2008
Deadline for contributions to the proceedings: Friday June 20th, 2008
Date of the workshop: Wednesday July 9th, 2008