单击此处编辑母版标题样式,编辑母版文本样式第二级第三级第四级第五级,9/18/18,#,人工智能及互联的未来课件,1,人工智能及互联的未来,AI and Our Interlinked Future,Tom,M.Mitchell,机器学习系,卡内基梅隆大学,匹兹堡,，,宾夕法尼亚洲,，,美国,人工智能及互联的未来,2,准确率,human,视觉,语音,human,准确率,机器人,准确率 human视觉语音 human准确率 机器,3,主题,:,机器学习,聊天机器人,recommendations,fraud detection,医疗诊断,机器翻译,深层神经网络,支持向量机,贝叶斯网络,强化学习,决策树,高斯混合模型,最大期望算法,.,主题:机器学习聊天机器人recommendationsfr,4,万物互联,-,人机共生,物联网,传感器,显示,会话界面,人类,机器,合作,机器作为可变通的队友,根据人类需求定制,重要的研究,问题,Critical Research Needed,持续的,永不停止的机器,学习,Continuous,Never-Ending Learning,可指导的机器,每个用户都是一个,老师,Instructable machines,every user a teacher,Internet of Things,Sensors,Displays,Conversational interfaces,Human,Machine Teams,Machines as flexible team mates,Customize to human user,万物互联 -人机共生物联网人类 机器 合作重要的研,5,当晚上下雨的时候,提前,30,分钟叫醒我,.,我不明白,。,可以教我吗,?,可指导的机器人,每个用户都是一个老师,/,程序员,Whenever it snows at night,wake me up 30 minutes earlier.,I dont understand.,Can you teach me?,当晚上下雨的时候,提前30分钟叫醒我.我不明白。可以教,6,我不明白,.,可以教我吗,?,告诉我的项目小组我们下午,4,点会议,Tell,my project team we will meet at 4pm.,好的,首先打开一个新的邮件,，,然后将收件人设为项目小组的所有组员,，,然后将邮件,i,题目设为,“,我们将在下午,4,点开会,，,然后发送,，,完毕,.,OK,first create a new email.,Then set its recipient to the email address of everyone on my project team.,Then set its subject to We will meet at 4pm.,And send it.,Thats all.,好的,，,明白了,.,Got it.,好的,现在告诉我的秘书我下午,4,点有事,.,OK,now tell my assistant I will be busy at 4pm.,执行,“,告诉我的秘书我下午,4,点有事,.”,Executing“tell my assistant I will be busy at 4.”,I dont understand.Can you teach me?,我不明白.可以教我吗?告诉我的项目小组我们下午4点会议T,7,Technical Approach,技术方法,可指导机器学习,“,告诉碧姗我下午有事,.”,(createInstance email),(set email recipient,(get emailAddress Bishan),(set email subject“I am late”),(send email),参考,:“LIA:A Natural Language Programmable Personal Assistant,”,EMNLP,2018.,“Instructable Intelligent Personal Agent,”,AAAI,2016.,2.,从用户的指令中学习,:,新的执行程序,提高自然语言语义分析模型,2.Learn from user instruction:,new procedures,improved language parser,1.Semantically parse each sentence into executable code.,“Tell Bishan I am late.”,(createInstance email),(set email recipient,(get emailAddress Bishan),(set email subject“I am late”),(send email),1.,将每个句子进行语义分析并转成可执行的代码,Technical Approach,8,可持续的,永不停止的学习,Continuous,Never-Ending Learning,可持续的,永不停止的学习,9,永不停止的语言学习机器,(NELL),NELL,知识库的一部分,知识,理解文字的能力的,增长率,问题,:,学习如何读互联网,.,方法,:,永不停止地学习,24x7,since 2010.,结果,:,知识图谱涵盖超过,1,千万的推测事实,持续地增长理解文字的能力,以及准确率,知识库,Problem:,Learn to read the web.,Approach:,Never ending learning,24x7,since 2010.,Results,:Knowledge graph with 100M beliefs,Continuously growing reading skills,knowledge,永不停止的语言学习机器(NELL)N,10,必须从无标注的数据中学习,半监督学习,优化,一致性从而提高准确率,必须避免平坦的学习曲线,不断发明新的学习任务持续扩展表示,形式,必须阶段学习任务,从易到难,永不停止的语言学习机器,(NELL):,得,到的教训,Lessons Learned,参考,:“Never Ending Learning,”,CACM,2018.,“Coupled Semi-Supervised Learning for,Information Extraction,”,WSDM,2010.,Learn from mostly unlabeled data,semi-supervised learning,optimize,consistency,to obtain,correctness,A,void flat learning curve,continuously invent new learning tasks,continuously extend representations,Stage learning tasks,easy tasks first,then harder,永不停止的语言学习机器(NELL):得到的教训,11,展望新的人工智能产品,Possible new AI products,展望新的人工智能产品Possible new AI prod,12,未来可能的产品,:,会话,助理,Conversational Assistants,Alibaba Genie,Amazon Echo,Google Home,免提,免看,可感知的,持续可用,Hands free,eyes free,perceptive,always on,每家零售店的过,道,Every aisle in every retail store,每个设备,的每个用户界面,Every user interface to every appliance,接下来,添加用户的教学能力,:,“,你刚才听到的声音是我的狗在叫,.,当你听到我的狗叫时,播放我的声音说,冷静下来。,“,Next,add ability for users to,teach,it:,“The sound you just heard was my dog barking.,When you hear my dog bark,play my voice saying calm down.“,未来可能的产品:会话助理 Conversati,13,未来可能的产品,:,可学习的灯,泡,Learning Light Bulbs,带手机功能的灯泡,摄像头,扬声器,麦克风,无线连接,成本低,体积大无电池,永不停止地学习能力,可指导,Light bulbs with cell phone abilities,camera,speaker,microphone,wireless connectivity,at low cost,large volume,no battery,with never ending learning abilities,and instructable,未来可能的产品:可学习的灯泡 L,14,未来可能的产品,:,可学习的灯,泡,Learning Light Bulbs,例子,:,在医院房间里,.,学会识别事物,(,护士、药品、访客,),学习跟踪和预测行动,(,护士到达时进行脉搏检测,),学会识别需要帮助,(,病人跌倒,),学会将电子工作顺序与执行相匹配,(,错误,),学会任何其他护士,病人教的,，,为他们做的事,Example:in hospital room.,learn to recognize things(nurses,medicines,visitors),learn to track and predict actions(nurse arrives,takes pulse),learn to recognize need for help(patient falls),learn to match electronic work order to execution(errors),anything else the nurses,patients,teach,it to do for them,未来可能的产品:可学习的灯泡,15,未来可能的产品,:,不断学习,可教的设备,Continuously Learning,Instructable Equipment,每一个,公共巴士,电,梯,打印机,飞机引擎,例子,:,每辆救护车,感知,:,观察外面和里面主动,:,提供驾驶指导,医疗建议,记忆,:,每一个行程,每个病人,对话,:,免提,免看的智能帮助学习,:,哪些人员需要哪种帮助,自我反省,:,了解何时需要服务,Every,public bus,elevator,printer,airplane engine,Example:every,ambulance,perceptive:watching outside and inside,proactive:offer driving directions,medical advice,memory:of every ride,every patient,conversational:hands-free,eyes-free assistance,learning:which personnel need which kind of help,self-reflective:notices when it needs servicing,未来可能的产品:不断学习,可教的设备,16,对社会的影响,Implications for Society,对社会的影响Implications for Society,17,人工智能带来的机会,AI Opportunities,改善交通改善医疗保健,改善教育智能城市,改善老人生活,Imp