norman,martin,william

We consider approximate inference in hybrid Bayesian Networks (BNs) and present a new iterative algorithm that efficiently combines dynamic discretisation with robust propagation algorithms on junction trees structures. Our approach offers a significant extension to Bayesian Network theory and practice by offering a flexible way of modelling continuous nodes in BNs conditioned on complex configurations of evidence and intermixed with discrete nodes as both parents and children of continuous nodes. Our algorithm is implemented in a commercial Bayesian Network software package, AgenaRisk, which allows model construction and testing to be carried out easily. The results from the empirical trials clearly show how our software can deal effectively with different type of hybrid models containing elements of expert judgement as well as statistical inference. In particular, we show how the rapid convergence of the algorithm towards zones of high probability density, make robust inference analysis possible even in situations where, due to the lack of information in both prior and data, robust sampling becomes unfeasible.

Although Bayesian Nets (BNs) are increasingly being used to solve real world risk problems, their use is still constrained by the difficulty of constructing the node probability tables (NPTs). A key challenge is to construct relevant NPTs using the minimal amount of expert elicitation, recognising that it is rarely cost-effective to elicit complete sets of probability values. We describe a simple approach to defining NPTs for a large class of commonly occurring nodes (called ranked nodes). The approach is based on the doubly truncated Normal distribution with a central tendency that is invariably a type of weighted function of the parent nodes. In extensive real-world case studies we have found that this approach is sufficient for generating the NPTs of a very large class of nodes. We describe one such case study for validation purposes. The approach has been fully automated in a commercial tool, called AgenaRisk, and is thus accessible to all types of domain experts. We believe this work represents a useful contribution to BN research and technology since its application makes the difference between being able to build realistic BN models and not.

This paper shows how recent revolutionary Bayesian Network (BN) algorithms can be used to model very complex reliability problems in a simple unified way. The algorithms work for socalled hybrid BNs, which are BNs that can contain a mixture of both discrete and continuous variables. Such hybrid BNs enable us to model failure times and reliability together. The approach allows a compact representation of the event-dependent failure behaviours characteristic of fault-tolerant systems, avoiding the state space explosion problem of the Markov Chain based approaches. The BN framework presented is able to solve any configuration of static and dynamic gates with general time-to-failure distributions, without using numerical integration techniques or simulation methods. Unlike other approaches (which tend to be restricted to using exponential distributions) we can use as input any parametric or empirical failure rate distribution. The approach offers a powerful framework for analysts and decision makers to successfully perform robust reliability assessment. Sensitivity, uncertainty, diagnosis analysis, common cause failures, and warranty analysis can also be easily performed within this framework.

To make accurate predictions of attributes like defects found in complex software projects we need a rich set of process factors. We have developed a causal model that includes such process factors, both quantitative and qualitative. The factors in the model were identified as part of a major collaborative project. A challenge for such a model is getting the data needed to validate it. We present a dataset, elicited from 31 completed software projects in the consumer electronics industry, which we used for validation. The data were gathered using a questionnaire distributed to managers of recent projects. The dataset will be of interest to other researchers evaluating models with similar aims. We make both the dataset and causal model available for research use. 1.

Although a number of approaches have been taken to quality prediction for software, none have achieved widespread applicability. This paper describes a single model to combine the diverse forms of, often causal, evidence available in software development in a more natural and efficient way than done previously. We use Bayesian Networks as the appropriate formalism for representing defect introduction, detection and removal processes throughout any life-cycle. The approach combines subjective judgements from experienced project managers and available defect rate data to produce a risk map and use this to forecast and control defect rates. Moreover, the risk map more naturally mirrors real world influences without any distracting mathematical formality. The paper focuses on the extensive validation of the approach within Philips Consumer Electronics (dozens of diverse projects across Philips internationally). The resulting model (packaged within a commercial software tool, AgenaRisk, usable by project managers) is now being used to predict defect rates at various testing and operational phases. The results of the validation confirm that the approach is scalable, robust and more accurate that can be achieved using classical methods. We have found 95 % correlation between actual and predicted defects. The defect prediction models incorporate cutting-edge ideas and results from software metrics and process improvement research and package them as risk templates that can either be applied either offthe-shelf or after calibrating them to local conditions and to suit the software development processes in use. 1.

Abstract. This paper challenges the traditional approach for assessing the overall quality of a software product, which is based on the assumption that, in ISO/IEC 9126 terms, a good external quality ensures a good quality-in-use. Here we change the focus of the quality assessment, concentrating on the quality-in-use as the driving factor for designing a software product, or for selecting the product that better fits a user’s needs. We propose a “backwards ” analysis of the relationship between the external quality and the quality-in-use which tries to determine the external quality sub-characteristics that are really relevant to ensure the required level of quality in a given context of use, in order to avoid superfluous costs or irrelevant features – which may unnecessarily increase the final price of the product. In this paper we propose Bayesian Belief Networks to model such relationships, and propose a method to build them for different contexts of use.

This chapter provides an introduction to the use of Bayesian Network (BN) models in Software Engineering. A short overview of the theory of BNs is included, together with an explanation of why BNs are ideally suited to dealing with the characteristics and shortcomings of typical software development environments. This theory is supplemented and illustrated using real world models that illustrate the advantages of BNs in dealing with uncertainty, causal reasoning and learning in the presence of limited data.