When constructing a Bayesian network, it can be advantageous to employ structural learning algorithms to combine knowledge captured in databases with prior information provided by domain experts. Unfortunately, conventional learning algorithms do not easily incorporate prior information, if this information is too vague to be encoded as properties that are local to families of variables. For instance, conventional algorithms do not exploit prior information about repetitive structures, which are often found in object oriented domains such as computer networks, large pedigrees and genetic analysis.

Languages for open-universe probabilistic models (OUPMs) can represent situations with an unknown number of objects and identity uncertainty. While such cases arise in a wide range of important real-world applications, existing general purpose inference methods for OUPMs are far less efficient than those available for more restricted languages and model classes. This paper goes some way to remedying this deficit by introducing, and proving correct, a generalization of Gibbs sampling to partial worlds with possibly varying model structure. Our approach draws on and extends previous generic OUPM inference methods, as well as auxiliary variable samplers for nonparametric mixture models. It has been implemented for BLOG, a well-known OUPM language. Combined with compile-time optimizations, the resulting algorithm yields very substantial speedups over existing methods on several test cases, and substantially improves the practicality of OUPM languages generally. 1

Because the advantages and limitations of white-box and black-box models are complementary, their ecient combination would give rise to superior models that achieve both the prior knowledge incorporation of white-box models and the data parsimony of blck-box models. We describe here a hybrid methodology for the representation and formalization of domain knowledge in a belief network and show how this formalized knowledge can enhance the learning and inference of a Bayesian multilayer perceptron. Firstly, we link the formalized probabilistic domain knowledge to the information sources through annotations (annotated belief network) to support the development of complex knowledge models. Secondly, we present two techniques to derive an informative prior for a Bayesian multilayer perceptron from a belief network. Thirdly, we show how to use belief networks for the management of missing values in black-box models. We report results on a real-world medical task: the prediction of the malignancy of ovarian masses from clinical measurements. The learning curve of the Bayesian multilayer perceptron with informative prior is better than those of the belief network and of the multilayer perceptron, and it illustrates that our technique can be advantageous in many situations.

This dissertation is devoted to the design of an intelligent, distributed fault and performance management system for communication networks. The architecture is based on a distributed agent paradigm, with belief networks as the framework for knowledge representation and evidence propagation. The dissertation consists of four major parts. First, we choose the mobile code technology to help implement a distributed, extensible framework for supporting adaptive, dynamic network monitoring and control. The focus of our work is on three aspects. First, the design of the standard infrastructure, or Virtual Machine, based on which agents could be created, deployed, managed and initiated to run. Second, the collection API for our delegated agents to collect data from network elements. Third, the callback mechanism through whichthe functionality of the delegated agents or even the native software could be extended. We propose three system designs based on such ideas. Second, we propose a distributed framework for intelligent fault management purpose. The managed network is divided into several domains and for each

BLOG is a powerful language to express models with an unknown number of objects and identity uncertainty. Current inference engines for BLOG are either too slow or require users to write a model-specific proposal distribution. We describe here, ongoing work to design a new, fast, generic inference engine for BLOG called blogc. The new implementation uses Gibbs sampling for finite-valued variables and performs an analysis of the model to generate customized sampling code in C. We describe our algorithms and methods in the context of various commonly used models and demonstrate significant performance improvement. 1.