Hiroshi FujiwaraWesometimeshavetomakeadecisionimmediately,withnoinformationaboutthefuture.Forexample,“whichisthebetter,torentortobuyaskigear?"Or,“howtopackloadsthatarearrivingoneafteranotherintotrucks?"Ifyouknewaboutthefuture,youcouldmakeagoodchoiceexploitingthatinformation.But,itisoftenthecasethat"nooneknowswhatmayhappentomorrow."Inourlaboratory,werevealthenatureunderlyingsuchsituationsandstudystrategiesandalgorithmsthatrunwithoutfutureinformation.Ouraimistodesignstrategiesandalgorithmsthatoutputthe"bestpossiblechoice"foreachmomentunderthesettingthatonecanhavehopelesslylittleinformationaboutthefuturelike"nooneknowswhatmayhappentomorrow."Furthermore,weclarifythenatureoftheinherentvalueofinformation,bytheoreticallycomparingthecaseswithorwithoutfutureinformation.AssociateProfessorPh.D. from Kyoto Univ. During his career, he visited Univ. Freiburg and UESTC, China. His research interests include algorithms and optimization.The "ski-rental problem" is equivalent to the choice of a strategy of automatically turning off the backlight on a mobile phone.ManagingtheUncertaintyoftheFutureThe "bin-packing problem" has application in several areas such as logistics or distributed systems.Inourlaboratorythestudentsacquireandpolishuptheircomprehensiveskilltosolveproblems.Specifically,theskillconsistsof(1)tograspthenatureunderlyingtheproblemprecisely,and(2)toconsiderwhattoolsorapproachesleadustoasolution.After GraduationIn the FutureElectrical& Computer EngineeringElectrical& Computer EngineeringIn the FutureAfter GraduationMinoru MaruyamaMultimodalsiamesenetworkarchitecturewhichembedsimage-basedandtext-basedfeaturesintoaunifiedvectorspaceAnexampleofimagesearchbasedonmultimodalanalogy.InMaruyamaLab.,ourresearchfocusesonmachinelearningtechniquesforvisualinformationunderstanding&processing.We,human,caneasilyrecognizeandunderstandthecontentsincludedintheimage.Thisabilityisimprovedthroughlearningfromoureverydayexperience.Ourresearchgoalistoprovidingsuchsignificantabilitytocomputer.Forthatpurposewemainlyinvestigatingdeeplearningtechniquesforimageunderstanding,contentbasedimagesearch,patternrecognition,3Drecoveryfromimages,andcomputergraphics.We are investigating machine learning techniques for visual information understanding. Especially we are studying deep learning techniques for acquiring appropriate latent representations. Our research interest also include computer graphics such as photo-realistic rendering algorithms. These techniques can be used for various kinds of applications including content based image query on the web, human-computer interaction. Over half of the graduates go to graduate school (master program of ShinshuUniversity).Graduates of our lab are employed by electronics companies, IT companies, including video game companies etc. Many graduates are working as software/IT engineers. ProfessorSince 1996, he is with ShinshuUniv. Before joining Shinshu Univ., he was with Mitsubishi Electric Corp. 1990-1991, he was a visiting scientist of Artif. Intell. Lab., MIT Learning Latent Representation for VisualInformation Understanding32
元のページ ../index.html#34