Probabilistic Graphical Models: Principles and Applications PDF
Preview Probabilistic Graphical Models: Principles and Applications
Advances in Computer Vision and Pattern Recognition Luis Enrique Sucar Probabilistic Graphical Models Principles and Applications Second Edition Advances in Computer Vision and Pattern Recognition Founding Editor Sameer Singh, Rail Vision, Castle Donington, UK Series Editor Sing Bing Kang, Zillow, Inc., Seattle, WA, USA Advisory Editors Horst Bischof, Graz University of Technology, Graz, Austria Richard Bowden, University of Surrey, Guildford, Surrey, UK Sven Dickinson, University of Toronto, Toronto, ON, Canada Jiaya Jia, The Chinese University of Hong Kong, Shatin, New Territories, Hong Kong Kyoung Mu Lee, Seoul National University, Seoul, Korea (Republic of) Yoichi Sato, University of Tokyo, Tokyo, Japan Bernt Schiele, Max Planck Institute for Informatics, Saarbrücken, Saarland, Germany Stan Sclaroff, Boston University, Boston, MA, USA More information about this series at http://www.springer.com/series/4205 Luis Enrique Sucar Probabilistic Graphical Models Principles and Applications Second Edition 123 LuisEnrique Sucar Instituto Nacional deAstrofísica, Ópticay Electrónica (INAOE) SanAndrés Cholula, Puebla,Mexico ISSN 2191-6586 ISSN 2191-6594 (electronic) Advances in Computer Vision andPattern Recognition ISBN978-3-030-61942-8 ISBN978-3-030-61943-5 (eBook) https://doi.org/10.1007/978-3-030-61943-5 1stedition:©Springer-VerlagLondon2015 2ndedition:©SpringerNatureSwitzerlandAG2021 Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpart of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission orinformationstorageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilar methodologynowknownorhereafterdeveloped. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publicationdoesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfrom therelevantprotectivelawsandregulationsandthereforefreeforgeneraluse. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained hereinorforanyerrorsoromissionsthatmayhavebeenmade.Thepublisherremainsneutralwithregard tojurisdictionalclaimsinpublishedmapsandinstitutionalaffiliations. ThisSpringerimprintispublishedbytheregisteredcompanySpringerNatureSwitzerlandAG Theregisteredcompanyaddressis:Gewerbestrasse11,6330Cham,Switzerland To my family, Doris, Edgar and Diana, for their unconditional love and support. Foreword Probabilistic graphical models (PGMs), and their use for reasoning intelligently under uncertainty, emerged in the 1980s within the statistical and artificial intelli- gence reasoning communities. The Uncertainty in Artificial Intelligence (UAI) conference became the premier forum for this blossoming research field. It was at UAI-92 in San Jose that I first met Enrique Sucar—both of us graduate students—where he presented his work on relational and temporal models for high-levelvisionreasoning.Enrique’simpressiveresearchcontributionstoourfield over the past 25 years have ranged from the foundational work on objective probabilities, to developing advanced forms of PGMS such as temporal and event Bayesiannetworks,tothelearningofPGMs,forexample,hismorerecentworkon Bayesian chain classifiers for multi-dimensional classification. Probabilistic graphical models are now widely accepted as a powerful and mature technology for reasoning under uncertainty. Unlike some of the ad hoc approaches taken in early experts systems, PGMs are based on the strong mathe- matical foundations of graph and probability theory. They can be used for a wide range of reasoning tasks including prediction, monitoring, diagnosis, risk assess- ment and decision making. There are many efficient algorithms for both inference and learning available in open-source and commercial software. Moreover, their power and efficacy have been proven through their successful application to an enormous range of real-world problem domains. Enrique Sucar has been a leading contributor in this establishment of PGMs as practical and useful technology, with his work across a wide range of application areas. These include medicine, reha- bilitationandcare,roboticsandvision,education,reliabilityanalysisandindustrial applications ranging from oil production to power plants. The first authors to drawn upon the early research on Bayesian networks and craft it into compelling narratives in the book form were Judea Pearl in Probabilistic Reasoning in Intelligent Systems and Rich Neapolitan in ProbabilisticReasoninginExpertSystems.ThismonographfromEnriqueSucaris a timely addition to the body of literature following Pearl and Neapolitan, with up-to-datecoverageofabroaderrangeofPGMsthanotherrecenttextsinthisarea: various classifiers, hidden Markov models, Markov random fields, Bayesian vii viii Foreword networks anditsdynamic, temporalandcausal variants,relationalPGMs,decision graphs and Markov decision process. It presents these PGMs, and the associated methodsforreasoning(orinference)andlearning,inaclearandaccessiblemanner, makingitsuitableforadvancedstudentsaswellasresearchersorpractitionersfrom other disciplines interested in using probabilistic models. The text is greatly enri- ched by the way Enrique has drawn upon his extensive practical experience in modelling with PGMs, illustrating their use across a diverse range of real-world applications from bioinformatics to air pollution to object recognition. I heartily congratulate Enrique on this book and commend it to potential readers. Melbourne, Australia Ann E. Nicholson May 2015 Preface Highlights of the Second Edition (cid:129) A new chapter on Partially Observable Markov Decision Process has been incorporated, which includes a detailed introduction to these models, approxi- mate solution techniques and application examples. (cid:129) Thechapteroncausalmodelshasbeenextendedanddividedintotwochapters, oneoncausalgraphicalmodels,includingcausalinference,andotheroncausal discovery. Several causal discovery techniques are presented, and additional application examples. (cid:129) Itincludesanewchapterthatgivesanintroductiontodeepneuralnetworksand their relation with probabilistic graphical models; presenting an analysis of different schemes for integrating deep neural networks and probabilistic graphicalmodels,andsomeexamplesoftheapplicationofthesehybridmodels in different domains. (cid:129) Additional types of classifiers are described, including Gaussian Naive Bayes, CircularChainClassifiers,andHierarchicalClassifierswithBayesianNetworks. (cid:129) The chapter on Hidden Markov Models incorporates Gaussian Hidden Markov Models. (cid:129) A knowledge transfer scheme for learning Bayesian networks is described. (cid:129) SamplingtechniquesforDynamicBayesianNetworksincludingParticleFilters have been added. (cid:129) It incorporates an additional method to solve Influence Diagrams based on the transformation to a Decision Tree. (cid:129) Several additional application examples have been incorporated. (cid:129) The number of problems in each chapter has been increased by 50%. (cid:129) APythonLibraryforinferenceandlearningforProbabilisticGraphicalModels has been developed, implementing several of the algorithms described in the book. ix x Preface Overview Probabilistic graphical models have become a powerful set of techniques used in several domains. This book provides a general introduction to probabilistic graphical models (PGMs) from an engineering perspective. It covers the funda- mentalsofthemainclassesofPGMs:Bayesianclassifiers,hiddenMarkovmodels, Bayesian networks, dynamic and temporal Bayesian networks, Markov random fields, influence diagrams, Markov decision processes and partially observable Markov decision processes; including representation, inference and learning prin- ciplesforeachone.ItdescribesseveralextensionsofPGMs:relationalprobabilistic models, causal models, and hybrid models. Realistic applications for each type of model are covered in the book. Some key features are the following: (cid:129) ThemainclassesofPGMsarepresentedinasinglemonographunderaunified framework. (cid:129) Thebookcoversthefundamentalaspects:representation,inferenceandlearning for all the techniques. (cid:129) Itillustratestheapplicationofthedifferent techniques inpracticalproblems,an important feature for students and practitioners. (cid:129) It includes some of the latest developments in the field, such as multidimen- sional and hierarchical Bayesian classifiers, relational graphical models, causal models and causal discovery, and hybrid deep neural networks-graphical models. (cid:129) Each chapter has a set of exercises, including suggestions for research and programming projects. Motivating the application of probabilistic graphical models to real-world problemsisoneofthegoals ofthis book.Thisrequires notonly knowledge ofthe different models and techniques, but also some practical experience and domain knowledge. To help the professionals in different fields gain some insight into the use of PGMs for solving practical problems, the book includes many examples of the application of the different types of models in a wide range of domains, including: (cid:129) Computer vision. (cid:129) Biomedical applications. (cid:129) Industrial applications. (cid:129) Information retrieval. (cid:129) Intelligent tutoring systems. (cid:129) Bioinformatics. (cid:129) Environmental applications. (cid:129) Robotics.
The list of books you might like
