Causality, Correlation and Artificial Intelligence for Rational Decision Making

Transcription

1 Causality, Correlation and Artificial Intelligence for Rational Decision Making

2 This page intentionally left blank

3 Causality, Correlation and Artificial Intelligence for Rational Decision Making Tshilidzi Marwala University of Johannesburg, South Africa World Scientific NEW JERSEY LONDON SINGAPORE BEIJING SHANGHAI HONG KONG TAIPEI CHENNAI

4 Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore USA office: 27 Warren Street, Suite , Hackensack, NJ UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE Causality, Correlation and Artificial Intelligence for Rational Decision Making Downloaded from Library of Congress Cataloging-in-Publication Data Marwala, Tshilidzi, 1971 Causality, correlation, and artificial intelligence for rational decision making / by Tshilidzi Marwala (University of Johannesburg, South Africa). pages cm Includes bibliographical references and index. ISBN (alk. paper) 1. Decision making. 2. Artificial intelligence. I. Title. T57.95.M '51--dc British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. Copyright 2015 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the publisher. For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher. In-house Editors: Sutha Surenddar/Chelsea Chin Typeset by Stallion Press enquiries@stallionpress.com Printed in Singapore

5 Preface Causality, correlation and artificial intelligence for rational decision making covers five significant areas of intellectual inquiry which are causality, correlation, artificial intelligence, rationality, and decision making. It gives new insights on causality and correlations machines and proposes that these form as a fundamental unit for rational decision making. Causality has been a subject of study for many years and has often been confused with correlation. Human intuition has evolved such that it has learned to identify causality through correlation. This book proposes a model that claims that within any causality model is buried a correlation machine. This study uses artificial intelligence to build correlation and causality machines. Artificial intelligence methods that are used are the multi-layer perceptron (MLP), radial basis function (RBF), fuzzy inference systems, support vector machines, genetic algorithms (GAs), and simulated annealing (SA). The concepts of causality, correlation and artificial intelligence are then applied in interstate conflict, condition monitoring and several biomedical applications. This book introduces the concept of flexibly-bounded rationality and uses the correlation and causal machines to implement this. Flexibly-bounded rationality is a special case of bounded rationality. Furthermore, this book introduces the theory of the marginalization of irrationality for decision making and applies this to the problem of breast cancer diagnosis. This book is intended for philosophers, computer scientists, engineers, and other interested parties at both undergraduate and graduate levels as well as practitioners. v

6 This page intentionally left blank

7 Acknowledgments I would like to thank the following institutions for contributing toward the writing of this book: University of Cambridge, University of Pretoria, and University of Johannesburg. I also would like to thank my students for their assistance in developing this manuscript more particularly Fholisani Mashegana. I dedicate this book to the schools that gave me the foundation to always seek excellence in everything I do and these are: Mbilwi Secondary School, Case Western Reserve University, University of Pretoria, University of Cambridge (St. John s College), and Imperial College (London). I thank the Stellenbosch Institute of Advanced Study (STIAS) for the financial support as well as accommodation during the writing of this book. This book is dedicated to the following people: Nhlonipho Khathutshelo, Lwazi Thendo, Mbali Denga Marwala as well as Dr. Jabulile Vuyiswa Manana. Professor Tshilidzi Marwala, Ph.D. Johannesburg 1 February 2014 vii

8 This page intentionally left blank

9 Preface Acknowledgments Contents 1. Introduction to Artificial Intelligence based Decision Making Introduction Correlation What is correlation? Correlation function Causality What is causality? Theories of causality What is a causal function? How to detect causation? Introduction to Artificial Intelligence Neural networks Hopfield networks Genetic algorithm Particle swarm optimization Simulated annealing Rational Decision Making Summary and Outline of the Book Conclusions References v vii 2. What is a Correlation Machine? Introduction Correlation Machines Auto-associative memory network ix

10 x Contents Causality, Correlation and Artificial Intelligence for Rational Decision Making Downloaded from Principal component analysis Expectation maximization algorithm Genetic Algorithm Initialization Crossover Mutation Reproduction Termination Multi-layer Perceptron Experimental Comparison Conclusions References What is a Causal Machine? Introduction Induction, Deduction, and Abduction What is Causality? Multi-layer Perceptron Causal Machine The architecture of the MLP causal machine Interstate conflict Radial Basis Function Causal Machine Theoretical foundation Applications to condition monitoring Fuzzy Inference System Causal Machine Theoretical foundation Application to a steam generator Conclusions References Correlation Machines Using Optimization Methods Introduction Multi-layer Perceptron Neural Network Missing Data Estimation Technique Genetic Algorithms Initialization Crossover Mutation Reproduction Termination... 70

11 Contents xi Causality, Correlation and Artificial Intelligence for Rational Decision Making Downloaded from Particle Swarm Optimization Simulated Annealing SA parameters Transition probabilities Monte Carlo method Markov Chain Monte Carlo Acceptance probability function: Metropolis algorithm Cooling schedule Missing Data Estimation: Case Studies Mechanical system Modeling of beer tasting Conclusions References Neural Networks for Modeling Granger Causality Introduction Granger Causality Multi-layer Perceptron for Granger Causality Bayesian statistics Hybrid Monte Carlo (HMC) RBF for Granger Causality The k-means Pseudo-inverse methods Example: Mackey Glass System Conclusions References Rubin, Pearl and Granger Causality Models: A Unified View Introduction Neyman-Rubin Causal Model Missing data mechanism Missing data imputation methods Pearl Causality Directed acyclic graph Associations between variables d-separation Back-door adjustment Front-door adjustment

12 xii Contents Causality, Correlation and Artificial Intelligence for Rational Decision Making Downloaded from Rules for do-calculus Pearl inferred causation algorithm Examples of using do-calculus Granger Causality Comparison: Neyman-Rubin, Pearl and Granger Causality Conclusions References Causal, Correlation and Automatic Relevance Determination Machines for Granger Causality Introduction Causal Machine to Granger Causality Multi-layer perceptron Scaled conjugate gradient method Correlation Machine to Granger Causality Auto-associative network for missing data estimation Nelder-Mead simplex optimization method Granger causality Automatic Relevance Determination for Granger Causality Experimental Investigation: Mackey Glass Time-Delay Differential Equation Conclusions References Flexibly-bounded Rationality Introduction Rational Decision Making: A Causal Approach Rational Decision Making Process Bounded-Rational Decision Making Flexibly-bounded Rational Decision Making Advanced information processing Missing data estimation Intelligent machines Experimental Investigations Condition monitoring HIV modeling Conclusions References

13 Contents xiii Causality, Correlation and Artificial Intelligence for Rational Decision Making Downloaded from 9. Marginalization of Irrationality in Decision Making Introduction Rational Decision Making What is Irrationality? Marginalization of Irrationality Theory Irrational Decision Making and the Theory of Marginalization of Irrationality in Decision Making Application of the Marginalization of Irrationality Theory for Breast Cancer Diagnosis MLP RBF Auto-associative neural network based on the MLP Auto-associative network based on the RBF Conclusions References Conclusions and Further Work Introduction Way Forward References Index 191