We consider the problem of model selection and accounting for model uncertainty in high-dimensional contingency tables, motivated by expert system applications. The approach most used currently is a stepwise strategy guided by tests based on approximate asymptotic P-values leading to the selection of a single model; inference is then conditional on the selected model. The sampling properties of such a strategy are complex, and the failure to take account of model uncertainty leads to underestimation of uncertainty about quantities of interest. In principle, a panacea is provided by the standard Bayesian formalism which averages the posterior distributions of the quantity of interest under each of the models, weighted by their posterior model probabilities. Furthermore, this approach is optimal in the sense of maximising predictive ability. However, this has not been used in practice because computing the posterior model probabilities is hard and the number of models is very large (often greater than 1011). We argue that the standard Bayesian formalism is unsatisfactory and we propose an alternative Bayesian approach that, we contend, takes full account of the true model uncertainty byaveraging overamuch smaller set of models. An efficient search algorithm is developed for nding these models. We consider two classes of graphical models that arise in expert systems: the recursive causal models and the decomposable

eflecting the importance or precision of dissimilarity # i j . 1. SOURCES OF DISTANCE DATA Dissimilarity information about a set of objects can arise in many different ways. We review some of the more important ones, organized by scientific discipline. 1.1. Geodesy. The most obvious application, perhaps, is in sciences in which distance is measured directly, although generally with error. This happens, for instance, in triangulation in geodesy. We have measurements which are approximately equal to distances, either Euclidean or spherical, depending on the scale of the experiment. In other examples, measured distances are less directly related to physical distances. For example, we could measure airplane or road or train travel distances between different cities. Physical distance is usually not the only factor determining these types of dissimilarities. 1 2 J. DE LEEUW<

Abstract not found

Abstract not found

We address the problem of updating a probability distribution represented by a Bayesian network upon presentation of soft evidence. Our motivation

How should flows through a network be organized, so that the network responds sensibly to failures and overloads? The question is currently of considerable technological importance in connection with the development of computer and telecommunication networks, while in various other forms it has a long history in the fields of physics and economics. In all of these areas there is interest in how simple, local rules, often involving random actions, can produce coherent and purposeful behaviour at the macroscopic level. This paper describes some examples from these various fields, and indicates how analogies with fundamental concepts such as energy and price can provide powerful insights into the design of routing schemes for communication networks.

Abstract not found

Rather than collect data from a variety of surveys, it is often more efficient to merge information from administrative lists. Matching of person files might be done using name and date-of-birth as the primary identifying information. There are obvious difficulties with entities having a commonly occurring name such as John Smith that may occur 30,000+ times (1.5 for each date-of-birth). If there are 5 % typographical error in each field, then using fast character-by-character searches can miss 20 % of true matches among noncommonly occurring records where name plus date-ofbirth might be unique. This paper describes some existing solutions and current research directions.

AbstractÐIn this paper, a novel method of pattern discovery is proposed. It is based on the theoretical formulation of a contingency table of events. Using residual analysis and recursive partitioning, statistically significant events are identified in a data set. These events constitute the important information contained in the data set and are easily interpretable as simple rules, contour plots, or parallel axes plots. In addition, an informative probabilistic description of the data is automatically furnished by the discovery process. Following a theoretical formulation, experiments with real and simulated data will demonstrate the ability to discover subtle patterns amid noise, the invariance to changes of scale, cluster detection, and discovery of multidimensional patterns. It is shown that the pattern discovery method offers the advantages of easy interpretation, rapid training, and tolerance to noncentralized noise. Index TermsÐPattern discovery, residual analysis, recursive partitioning, events, contingency tables.

This paper describes and analyzes a method for finding nontrivial solutions of the inequality Ax 0, where A is an m \Theta n matrix of rank n. The method is based on the observation that a certain function f has a unique minimum if and only if the inequality