ebook img

Regularization Theory for Ill-posed Problems: Selected Topics PDF

304 Pages·2013·1.879 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Regularization Theory for Ill-posed Problems: Selected Topics

Inverse and Ill-Posed Problems Series 58 ManagingEditor SergeyI.Kabanikhin,Novosibirsk,Russia;Almaty,Kazakhstan Shuai Lu, Sergei V. Pereverzev Regularization Theory for Ill-posed Problems Selected Topics De Gruyter MathematicsSubjectClassification2010 47A52,65J10,65J20,65J22,65N15,65N20 Authors Dr.ShuaiLu FudanUniversity SchoolofMathematicalSciences No.220,RoadHandan 200433Shanghai PRChina [email protected] Prof.Dr.SergeiV.Pereverzev AustrianAcademyofSciences JohannRadonInstituteforComputationalandAppliedMathematics(RICAM) Altenbergerstraße69 4040Linz Austria [email protected] ISBN978-3-11-028646-5 e-ISBN978-3-11-028649-6 Set-ISBN978-3-11-028650-2 ISSN1381-4524 LibraryofCongressCataloging-in-PublicationData ACIPcatalogrecordforthisbookhasbeenappliedforattheLibraryofCongress. BibliographicinformationpublishedbytheDeutscheNationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailedbibliographicdataareavailableintheInternetathttp://dnb.dnb.de. © 2013WalterdeGruyterGmbH,Berlin/Boston Typesetting:PTP-BerlinProtago-TEX-ProductionGmbH,www.ptp-berlin.de Printingandbinding:Hubert&Co.GmbH&Co.KG,Göttingen Printedonacid-freepaper PrintedinGermany www.degruyter.com ThisbookwaswrittenwithloveforAnna althoughitwasnotintendedtobeherfavoriteone. Thebookisjustevidencethathereffortswerenotuseless. AndtoYanqunforherconsistentsupport. Preface Thetheoryofinverseproblemshasawidevarietyofapplicationsbecauseanymath- ematicalmodelneedstobecalibratedbeforeitcanbeused,andsuchacalibrationis atypicalinverseproblem. Regularizationtheory,inturn,isthealgorithmicpartofthetheoryofinverseprob- lems. It provides and analyzes the methods for dealing with ill-posedness, which is oneofthemainissuesforinverseproblems. Inspiteofagrowingnumberofmonographsonregularizationtheory(atthetimeof writing,thelatestpublishedoneis[84]),therearequiteafewtopicswhichhaveonly recentlybeendevelopedandwhicharenotyetreflectedintheliterature.Thepresent bookismotivatedbysomeofthese. Thefirstnoveltyofthisbookisthatitsimultaneouslyanalyzestheill-posedprob- lems with deterministic and stochastic data noises. Not only does such analysis al- low uniform theoretical justification of a general regularization scheme for both of theabove-mentionednoisemodels,italsoprovidesalinktoalargeclassoflearning theoryalgorithms,whichareessentiallyallofthelinearregularizationschemes. Notethatthechapteronregularizationalgorithmsinlearningtheoryisanotherfea- ture which distinguishes this book from existing monographic literature on inverse problems. A further novelty of the book is Chapter 3, entitled “Multiparameter regulariza- tion”.Itisinterestingtoobservethatinexistingpublicationstheperformanceofmulti- parameter regularization schemes have been variously judged by authors. Some of them found that multiparameter regularization only marginally improved the one- parameterversion,whileothersreportedonmostsatisfactorydecisionsgivenbymulti- parameteralgorithmsincaseswheretheirone-parametercounterpartsfailed.Wehope thatChapter3willshedlightonthisslightlycontroversialsubject. Note that in this book the term “multiparameter regularization” is used as a syn- onym for “multiple penalty regularization”. At the same time, in modern numerical analysis,theapproximationandregularizationalgorithmsarebecomingmoresophis- ticatedanddependentonvariousparametersparameterizingeventhespaceswherepe- nalization,orregularization,isperformed.Theself-tuningofsuchparametersmeans that a regularization space is automatically adjusted to the considered problem. On theotherhand,classicalregularizationtheoryrestrictsitselftostudyingthesituation wherearegularizationspaceisassumedtobegivenapriori.Therefore,tothebestof ourknowledge,Chapter5ofthepresentbookisthefirstattemptinthemonographic literature to analyze the adaptive choice of the regularization space. This analysis is viii Preface basedontheconceptofmeta-learning,whichisalsofirstintroducedinthecontextof regularizationtheory. The meta-learning concept presupposes that the design parameters of algorithms are selected on the basis of previous experience with similar problems. Therefore, parameterselectionrulesdevelopedinthiswayareintrinsicallyproblem-oriented. InChapter5wedemonstrateameta-learning-basedapproachtoregularizationona problemfromdiabetestechnology,butitwillalsobeshownthatthemainingredients can be exploited in other applications. At the same time, the material of the chap- terdescribesoneofthefirstapplications ofregularizationtheoryindiabetes therapy management,whichisanextremelyimportantmedicalcarearea. Wehopethatsuchacontextmakesthebookofinterestforawideaudience,within theinverseproblemscommunityandbeyond. Thefirstpartofthebookcanalsoberecommendedforuseinlectures.Forexample, thesectionsoftheintroductorychaptercanbeusedindependentlyingeneralcourses onnumericalanalysisasexamplesofill-posedproblemsandtheirtreatments. Atthesametime,Chapter2containsacompactandrathergeneralpresentationof regularization theory for linear ill-posed problems in the Hilbert space setting under deterministicandstochasticnoisemodelsandgeneralsourceconditions.Thischapter canprovidematerialforanadvancedmastercourseonregularizationtheory.Sucha coursewasgivenattheTechnicalUniversityofKaiserslautern,Germany,andatthe Stefan Banach International Center in Warsaw, Poland. In Chapter 2 we have really triedtoadaptthepresentationforthispurpose.Forexample,weavoidedthenumer- ationoftheformulaeinordertomakethematerialmoreconvenientforpresentation ontheblackboard. Thesecondpartofthebookcanbeseenasapresentationofsomefurtherdevelop- mentsofthebasictheory.Thismaterialisnewinmonographicliteratureonregular- izationtheoryandcanbeusedinstudents’seminars. ThebookwaswritteninthestimulatingatmosphereoftheJohannRadonInstitute forComputationalandAppliedMathematics(RICAM).Thepreliminaryplanforthe projectwasdiscussedwithitsFoundingDirector,ProfessorHeinzEngl. Thebookwouldnotbepossiblewithoutourcolleagues.Weareespeciallygrateful toPeterMathé(WIAS,Berlin),AlexanderGoldenshluger(UniversityofHaifa),Eber- hardSchock(TechnicalUniversityofKaiserslautern),UlrichTautenhahn(12.01.1951 –10.10.2011),LorenzoRosasco(MIT,Cambridge),BerndHofmann(TechnicalUni- versityofChemnitz),HuiCao(SunYat-senUniversity),SivananthanSampath(IIT, Delhi), Valeriya Naumova (RICAM). We are also grateful to Christoph von Friede- burgandAnjaMöbiusfromDeGruyterforthefinalimpulsetostartwritingthisbook. SpecialthankstoJinCheng(FudanUniversity,Shanghai),whorecommendedthefirst authortothesecondoneasaPh.D.student,whichwasthebeginningofthestory. Finally we gratefully acknowledge financial support from the Austrian Fonds zur Förderung der wissenschaftlichen Forschung (FWF) (projects P17251-N12 and P20235-N18), Alexander von Humboldt Foundation, National Natural Science Preface ix FoundationofChina(KeyProjects91130004and11101093),ShanghaiScienceand Technology Commission (11ZR1402800 and 11PJ1400800) and the Programme of IntroducingTalentsofDisciplinetoUniversities(B08018),China. Shanghai–Linz ShuaiLuandSergeiV.Pereverzev December2012

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.