CONTENTS

Part I PROBABILISTIC REASONING

Chapter 1 Bayesian Reasoning

1.1 Reasoning under uncertainty

1.2 Uncertainty in AI

1.3 Probability calculus

1.3.1 Condiţional probability theorems

1.3.2 Variables

1.4 Interpretations of probability

1.5 Bayesian philosophy

1.5.1 Bayes' theorem

1.5.2 Betting and odds

1.5.3 Expected utility

1.5.4 Dutchbooks

1.5.5 Bayesian reasoning examples

1.6 The goal of Bayesian AI

1.7 Achieving Bayesian AI

1.8 Are Bayesian networks Bayesian?

1.9 Summary

1.10 Bibliographic notes

1.11 Technical notes

1.12 Problems

Chapter 2 Introducing Bayesian Networks

2.1 Introduction

2.2 Bayesian network basics

2.2.1 Nodes and values

2.2.2 Structure

2.2.3 Condiţional probabilities

2.2.4 The Markov property

2.3 Reasoning with Bayesian networks

2.3.1 Types of reasoning

2.3.2 Types of evidence

2.3.3 Reasoning with numbers

2.4 Understanding Bayesian networks

2.4.1 Representing the joint probability distribution

2.4.2 Pearl's network construction algorithm . . . .

2.4.3 Compactness and node ordering

2.4.4 Condiţional independence

2.4.5 d-separation

2.5 More examples

2.5.1 Earthquake

2.5.2 Metastatic cancer

2.5.3 Asia

2.6 Summary

2.7 Bibliographic notes

2.8 Problems

Chapter 3 Inference in Bayesian Networks

3.1 Introduction

3.2 Exact inference in chains

3.2.1 Two node network

3.2.2 Three node chain

3.3 Exact inference in polytrees

3.3.1 Kim and Pearl's message passing algorithm . .

3.3.2 Message passing example

3.3.3 Algorithm features

3.4 Inference with uncertain evidence

3.4.1 Using a virtual node

3.4.2 Virtual nodes in the message passing algorithm

3.5 Exact inference in multiply-connected networks ....

3.5.1 Clustering methods

3.5.2 Junction trees

3.6 Approximate inference with stochastic simulation . . .

3.6.1 Logic sampling

3.6.2 Likelihood weighting

3.6.3 Markov Chain Monte Carlo (MCMC)

3.6.4 Using virtual evidence

3.6.5 Assessing approximate inference algorithms . .

3.7 Other computations

3.7.1 Belief revision

3.7.2 Probability of evidence

3.8 Causal inference

3.9 Summary

3.10 Bibliographic notes

3.11 Problems

Chapter 4 Decision Networks

4.1 Introduction

4.2 Utilities

4.3 Decision network basics

4.3.1 Nodetypes

4.3.2 Football team example

4.3.3 Evaluating decision networks

4.3.4 Information links

4.3.5 Fever example

4.3.6 Types of actions

4.4 Sequential decision making

4.4.1 Test-action combination

4.4.2 Real estate investment example . . .

4.4.3 Evaluation using a decision tree model

4.4.4 Value of information

4.4.5 Direct evaluation of decision networks

4.5 Dynamic Bayesian networks

4.5.1 Nodes, structure and CPTs

4.5.2 Reasoning

4.5.3 Inference algorithms for DBNs . . . .

4.6 Dynamic decision networks

4.6.1 Mobile robot example

4.7 Summary

4.8 Bibliographic notes

4.9 Problems

Chapter 5 Applications of Bayesian Networks

5.1 Introduction

5.2 A brief survey of BN applications

5.2.1 Types of reasoning

5.2.2 BN stractures for medical problems .

5.2.3 Other medical applications

5.2.4 Non-medical applications

5.3 Bayesian poker

5.3.1 Five-card stud poker

5.3.2 A decision network for poker

5.3.3 Betting with randomization

5.3.4 Bluffing

5.3.5 Experimental evaluation

5.4 Ambulation monitoring and fall detection . .

5.4.1 The domain

5.4.2 The DBN model

5.4.3 Case-based evaluation

5.4.4 An extended sensor model

5.5 A Nice Argument Generator (NAG)

5.5.1 NAG architecture

5.5.2 Example: An asteroid strike

5.5.3 The psychology of inference

5.5.4 Example: The asteroid strike continues

5.5.5 The fu ture of argumentation

5.6 Summary

5.7 Bibliographic notes

5.8 Problems

Part II LEARNING CAUSAL MODELS

Chapter 6 Learning Linear Causal Models

6.1 Introduction

6.2 Path models

6.2.1 Wright's first decomposition rule

6.2.2 Parameterizing linear models

6.2.3 Learning linear models is complex

6.3 Condiţional independence learners

6.3.1 Markov equivalence

6.3.2 PCalgorithm

6.3.3 Causal discovery versus regression

6.4 Summary

6.5 Bibliographic notes

6.6 Technical notes

6.7 Problems

Chapter 7 Learning Probabilities

7.1 Introduction

7.2 Parameterizing discrete models

7.2.1 Parameterizing a binomial model

7.2.2 Parameterizing a multinomial model

7.3 Incomplete data

7.3.1 The Bayesian solution

7.3.2 Approximate Solutions

7.3.3 Incomplete data: summary

7.4 Learning local structure

7.4.1 Causal interaction

7.4.2 Noisy-or connections

7.4.3 Classification trees and graphs

7.4.4 Logit models

7.4.5 Dual model discovery

7.5 Summary

7.6 Bibliographic notes

7.7 Technical notes

7.8 Problems

Chapter 8 Learning Discrete Causal Structure

8.1 Introduction

8.2 Cooper & Herskovits' K2

8.2.1 Learning variable order

8.3 MDL causal discovery

8.3.1 Lam and Bacchus's MDL code for causal models

8.3.2 Suzuki's MDL code for causal discovery

8.4 Metric pattern discovery

8.5 CaMML: Causal discovery via MML

8.5.1 An MML code for causal structures

8.5.2 An MML metric for linear models

8.6 CaMML stochastic search

8.6.1 Genetic algorithm (GA) search

8.6.2 Metropolis search

8.6.3 Prior constraints

8.6.4 MML models

8.6.5 An MML metric for discrete models

8.7 Experimental evaluation

8.7.1 Qualitative evaluation

8.7.2 Quantitative evaluation

8.8 Summary

8.9 Bibliographic notes

8.10 Technical notes

8.11 Problems

Part III KNOWLEDGE ENGINEERING

Chapter 9 Knowledge Engineering with Bayesian Networks

9.1 Introduction

9.1.1 Bayesian network modeling tasks

9.2 The KEBN process . .

9.2.1 KEBN lifecycle model

9.2.2 Prototyping and spiral KEBN

9.2.3 Are BNs suitable for the domain problem? . . .

9.2.4 Process management

9.3 Modeling and elicitation

9.3.1 Variables and values

9.3.2 Graphical structure

9.3.3 Probabilities

9.3.4 Local structure

9.3.5 Variants of Bayesian networks

9.3.6 Modeling example: missing car

9.3.7 Decision networks

9.4 Adaptation

9.4.1 Adapting parameters

Summary

Appendix A Notation

Appendix B Software Packages

B.l Introduction

B.2 History

B.3 Murphy's Software Package Survey

B.4 BN software

B.4.1 Analytica

B.4.2 BayesiaLab

B.4.3 Bayes Net Toolbox (BNT)

B.4.4 GeNIe

B.4.5 Hugin

B.4.6 JavaBayes

B.4.7 MSBNx

B.4.8 Netica

B.5 Bayesian statistical modeling . . .

B.5.1 BUGS

B.5.2 First Bayes

B.6 Causal discovery programs ....

B.6.1 Bayesware Discoverer . .

B.6.2 CaMML

B.6.3 TETRAD

B.6.4 WinMine

References