人工智能中的深度学习——从机器感知到机器认知邓力.pptx

《人工智能中的深度学习——从机器感知到机器认知邓力.pptx》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、DeepLearningforAIfromMachinePerceptiontoMachineCognitionLiDengChiefScientistofAI,MicrosoftApplications/ServicesGroup(ASG)&MSRDeepLearningTechnologyCenter(DLTC)ThanksgotomanycolleaguesatDLTC&MSR,collaboratinguniversities,andatMicrosoft’sengineeringgroups(ASG+)

2、APlenaryPresentationatIEEE-ICASSP,March24,2016DefinitionDeeplearningisaclassofmachinelearningalgorithmsthat[1](pp199–200)•useacascadeofmanylayersofnonlinearprocessing.•arepartofthebroadermachinelearningfieldoflearningrepresentationsofdatafacilitatingend-to-en

3、doptimization.•learnmultiplelevelsofrepresentationsthatcorrespondtohierarchiesofconceptabstraction•…,…23Artificialintelligence(AI)istheintelligenceexhibitedbymachinesorsoftware.Itisalsothenameoftheacademicfieldofstudyonhowtocreatecomputersandcomputersoftwaret

4、hatarecapableofintelligentbehavior.Artificialgeneralintelligence(AGI)istheintelligenceofa(hypothetical)machinethatcouldsuccessfullyperformanyintellectualtaskthatahumanbeingcan.Itisaprimarygoalofartificialintelligenceresearchandanimportanttopicforsciencefictio

5、nwritersandfuturists.Artificialgeneralintelligenceisalsoreferredtoas"strongAI“…AI/(A)GI&DeepLearning:themainthesisAI/GI=machineperception(speech,image,video,gesture,touch...)+machinecognition(naturallanguage,reasoning,attention,memory/learning,knowledge,decis

6、ionmaking,action,interaction/conversation,…)GI:AIthatisflexible,general,adaptive,learningfrom1stprinciplesDeepLearning+Reinforcement/UnsupervisedLearningAI/GI4AI/GI&DeepLearning:howAlphaGofitsAI/GI=machineperception(speech,image,video,gesture,touch...)+machi

7、necognition(naturallanguage,reasoning,attention,memory/learning,knowledge,decisionmaking,action,interaction/conversation,…)AGI:AIthatisflexible,general,adaptive,learningfrom1stprinciplesDeepLearning+Reinforcement/UnsupervisedLearningAI/AGI5Outline•Deeplearni

8、ngformachineperception•Speech•Image•Deeplearningformachinecognition••••••SemanticmodelingNaturallanguageMultimodalityReasoning,attention,memory(RAM)Knowledgerepresentation/management/expl