Dear Patrick,

Along these very lines, you might find

P. J. Tan and D. L. Dowe (2003).  MML Inference of Decision Graphs
      with Multi-Way Joins and Dynamic Attributes, (to appear) In Proc.
      16th Australian Joint Conference on Artificial Intelligence (AI'03),
      Perth, Australia, 3-5 Dec. 2003

rather useful.

The notion of decision graph, with a disjunction (or join), originally
due to Jon Oliver and Chris Wallace in 1991 (that's the same Chris Wallace
who developed Minimum Message Length (MML) in 1968 and the Wallace
multiplier in 1964), is extended here to allow both multi-way joins and
dynamic attributes.

The disjunction resulting from a join can be regarded as a concept.
So, too, can the concept resulting from the new notion of a dynamic attribute.

There are empirical comparisons with C4.5 and C5 which, at least to me,
appear fairly clear.

The paper is downloadable from
http://www.csse.monash.edu.au/~dld/Publications/2003/Tan+Dowe2003.ref .


Regards,

David Dowe.



> From [log in to unmask] Fri Sep 19 00:48:39 2003
> Subject: Announcement of PhD position
> To: [log in to unmask]
>
> The Erasmus Research Institute of Management (ERIM) has a four-year position
> available for the following PhD project.
>
> Title:        Learning Concept Hierarchies in Multi-Attribute Decision
> Making
> Supervisors:  Prof. dr. P.J.F. Groenen ([log in to unmask]) and
>               Dr. J.C. Bioch ([log in to unmask])
> Affiliation:  Rotterdam School of Economics, Erasmus University Rotterdam,
>               The Netherlands
> Period:       Four years
> Salary:       approximately starting at euro 1500 in the first year
>               to euro 2100 in the fourth year
> URL:          http://www.fbk.eur.nl/ERIM/PHD/VACANCIES/mkt.html
>
> Summary:
> Large data sets with a large number of variables (or attributes) are getting
> more and more prevalent in economics. Therefore, structuring the variables
> also gains importance especially in the case of predicting an outcome
> variable. In this project, we focus on a promising new approach in
> artificial intelligence that has been proposed recently in the literature.
> The method searches for a hierarchy of variables in an automated fashion.
> This method of concept hierarchies can be graphically represented by a tree
> structure where each branch is split into mutually exclusive sets of
> predictor variables. Such a tree has the important advantage that it is easy
> to interpret from a substantive point of view. Another advantage is a higher
> accuracy in prediction. In this project, we consider study the use of
> concept hierarchies to multi-attribute decisions as they arise, for example,
> in consumer research and financial decision-making. Since the method of
> concept hierarchies is in its beginning stage, much of the method is not
> clear yet, for example, how to treat noisy or ordinal data. Therefore, the
> main objective of this project is to study, apply, and extend the method of
> concept hierarchies.
>
> We ask
> - Knowledge of matrix algebra and interest in mathematical modelling.
> - Some experience in data analysis and statistics.
> - Good knowledge of a modern low level programming language
>   (e.g., Visual Basic, JAVA, Delphi, Fortran, C++) or higher
>   level languages (e.g., MatLab, S-Plus, R, etc.).
> - Candidates should have a quantitative background
>   (for example, econometrics, statistics, psychometrics,
>   machine learning, or artificial intelligence).
> - Candidates with an economic background (for example, econometrics
>   or marketing) have an advantage.
>
> -------------------------------------------------
> Prof. dr. Patrick J.F. Groenen
> Econometric Institute
> Erasmus University Rotterdam
> Room H11.23
> P.O. Box 1738, 3000 DR Rotterdam, The Netherlands
> tel:    ++ 31 10 408 1281
> fax:    ++ 31 10 408 9162
> e-mail: [log in to unmask]
> -------------------------------------------------