人工智能课后习题答案部分已翻译考试.docx

人工智能课后习题答案部分已翻译考试.docx

《人工智能课后习题答案部分已翻译考试.docx》由会员分享，可在线阅读，更多相关《人工智能课后习题答案部分已翻译考试.docx（17页珍藏版）》请在冰豆网上搜索。

人工智能课后习题答案部分已翻译考试

1.1Defineinyourownword:

（a）intelligence,（b）artificialintelligence,（c）agent.

•Intelligence智能:

Dictionarydefinitionsofintelligencetalkabout“thecapacitytoacquireandapplyknowledge”or“thefacultyofthoughtandreason”or“theabilitytocomprehendandprofitfromexperience.”Theseareallreasonableanswers,butifwewantsomethingquantifiablewewouldusesomethinglike“theabilitytoapplyknowledgeinordertoperformbetterinanenvironment.”

智能的字典定义有一种学习或应用知识的能力，一种思考和推理的本领，领会并且得益于经验的能

力，这些都是有道理的答案，但如果我们想量化一些东西，我们将用到一些东西像为了在环境中更好的完成任务使能力适应知识

•Artificialintelligence人工智能:

Wedefineartificialintelligenceasthestudyandconstructionofagentprogramsthatperformwellinagivenenvironment,foragivenagentarchitecture.

作为一学习和构造智能体程序，为了一个智能体结构，在被给的环境中可以很好的完成任务。

•Agen智能体t:

Wedefineanagentasanentity实体thattakesactioninresponsetoperceptsfromanenvironment.在一个环境中对一个对象做出反应的实体

1.4Therearewell-knownclassesofproblemthatareintractablydifficultforcomputers,andotherclassesthatareprovablyundecidable.DoesthismeanthatAIisimpossible?

No.ItmeansthatAIsystemsshouldavoidtryingtosolveintractableproblems.Usually,thismeanstheycan

onlyapproximateoptimalbehavior.Noticethathumansdon’tsolveNPcompleteproblemseither.Sometimestheyaregoodatsolvingspecificinstanceswithalotofstructure,perhapswiththeaidofbackgroundknowledge.AIsystemsshouldattempttodothesame.

1.11“surelycomputerscannotbeintelligent-theycandoonlywhattheirprogrammerstellthem.”Isthelatterstatementtrue,anddoesitimplytheformer?

Thisdependsonyourdefinitionof“intelligent”and“tell.”Inonesensecomputersonlydowhatthe

programmerscommandthemtodo,butinanothersensewhattheprogrammersconsciouslytellsthecomputertodooftenhasverylittletodowithwhatthecomputeractuallydoes.Anyonewhohaswrittenaprogramwithanornerybugknowsthis,asdoesanyonewhohaswrittenasuccessfulmachinelearningprogram.SoinonesenseSamuel“told”thecomputer“learntoplaycheckersbetterthanIdo,andthenplaythatway,”butinanothersensehetoldthecomputer“followthislearningalgorithm”anditlearnedtoplay.Sowe’releftinthesituationwhereyoumayormaynotconsiderlearningtoplaycheckerstobessignofintelligence（oryoumaythinkthatlearningtoplayintherightwayrequiresintelligence,butnotinthisway）,andyoumaythinktheintelligenceresidesintheprogrammerorinthecomputer

Chapter2

2.1Defineinyourownwordsthefollowingterms:

agent,agentfunction,agentprogram,rationality,reflexagent,model-basedagent,goal-basedagent,utility-basedagent,learningagent.

Thefollowingarejustsomeofthemanypossibledefinitionsthatcanbewritten:

•Agent智能体:

anentity（实体）thatperceives（感知）andacts行为;or,onethatcanbeviewedasperceivingandacting.Essentially本质上anyobjectqualifies限定;thekeypointisthewaytheobject

implementsanagentfunction.（Note:

someauthorsrestrictthetermtoprogramsthatoperateonbehalfofahuman,ortoprogramsthatcancausesomeoralloftheircodetorunonothermachinesonanetwork,asinmobileagents.MOBILEAGENT）

一个具有感知和行文的实体，或者是一个可以观察到感觉的实体，本质上，任何限定对象，只要的观

点是一种对象执行智能体函数的方法。

（注意，一些作者）可以感知环境，并在环境中行动的某种东西。

•Agentfunction智能体函数:

afunctionthatspecifiestheagent’sactioninresponsetoeverypossibleperceptsequence.智能体相应任何感知序列所采取的行动

•Agentprogram智能体程序:

thatprogramwhich,combinedwithamachinearchitecture,implementsanagentfunction.Inoursimpledesigns,theprogramtakesanewperceptoneachinvocationandreturnsan

action.实现了智能函数。

有各种基本的智能体程序设计，反应出现实表现的一级用于决策过程的信息种类。

设计可能在效率、压缩性和灵活性方面有变化。

适当的智能体程序设计取决于环境的本性

•Rationality；理性:

apropertyofagentsthatchooseactionsthatmaximizetheirexpectedutility,giventheperceptstodate.

•Autonomy自主:

apropertyofagentswhosebehaviorisdeterminedbytheirownexperienceratherthansolelybytheirinitialprogramming.

•Reflexagent反射型智能体:

anagentwhoseactiondependsonlyonthecurrentpercept.

一个智能体的行为仅仅依赖于当前的知觉。

•Model-basedagent基于模型的智能体:

anagentwhoseactionisderiveddirectlyfromaninternalmodelofthecurrentworldstatethatisupdatedovertime.

一个智能体的行为直接得自于内在模型的状态，这个状态是当前世界通用的不断更新。

•Goal-basedagen基于目标的智能体t:

anagentthatselectsactionsthatitbelieveswillachieveexplicitlyrepresentedgoals.智能体选择它相信能明确达到目标的行动。

•Utility-basedagen基于效用的智能体t:

anagentthatselectsactionsthatitbelieveswillmaximizetheexpectedutilityoftheoutcomestate.试图最大化他们自己期望的快乐

•Learningagent学习智能体:

anagentwhosebehaviorimprovesovertimebasedonitsexperience.

2.2Boththeperformancemeasureandtheutilityfunctionmeasurehowwellanagentisdoing.Explainthedifferencebetweenthetwo.

Aperformancemeasure（性能度量）isusedbyanoutsideobservertoevaluate（评估）howsuccessfulan

agentis.Itisafunctionfromhistoriestoarealnumber.Autilityfunction（效用函数）isusedbyanagentitselftoevaluatehowdesirable（令人想要）statesorhistoriesare.Inourframework,theutilityfunction

maynotbethesameastheperformancemeasure;furthermore,anagentmayhavenoexplicitutilityfunctionatall,whereasthereisalwaysaperformancemeasure.

2.5Foreachoffollowingagents,developaPEASdescriptionofthetaskenvironment:

a.Robotsoccerplayer;

b.Internetbook-shoppingagent;

c.AutonomousMarsrover;

d.Mathematician’stheorem-provingassistant.

Somerepresentative,butnotexhaustive,answersaregiveninFigureS2.1.

智能体类型性能度量环境执行器传感器

机器人足球运动员因特网购

赢得比赛，打败对手获得请求/感兴趣

裁判，自己队伍，

其他队伍，自己身体装置（腿）行走踢球向下连接，输入提

相机，触摸传感器加速器

书智能体

的书，最小支出因特网

交数据，用户显示器网页，用户请求

自主火星漫步者数学家的定理证明助手

地形探测，汇报，样本采集分析

火星，运行装置，登陆器

轮子/腿，简单手机装置，分析装置，无线电发射装置

相机，触摸传感器方向传感器

2.6ForeachoftheagenttypeslistedinExercise2.5,characterizetheenvironmentaccordingtothepropertiesgiveninSection2.3,andselectasuitableagentdesign.Thefollowingexercisesallconcerntheimplementationofenvironmentandagentsforthevacuum-cleanerworld.

EnvironmentpropertiesaregiveninFigureS2.2.Suitableagenttypes:

a.Amodel-basedreflexagentwouldsufficeformostaspects;fortacticalplay,autilitybasedagentwithlookaheadwouldbeuseful.基于模型的映射能够满足大多数要求，对于战术游戏，向前效用智能体将会有用

b.Agoal-basedagentwouldbeappropriateforspecificbookrequests.Formoreopenendedtasks—e.g.,“Findmesomethinginterestingtoread”—tradeoffs（权衡折中）areinvolved（棘手的）andtheagentmustcompareutilitiesforvarious（不同的）possiblepurchases.基于目标的智能体将适当的明确书的请求，为更多开放的任务，例如查找我有兴趣读的书，智能体必须比较各种可能的购买方式之间的效用

c.Amodel-basedreflexagentwouldsufficeforlow-levelnavigationandobstacleavoidance;forroute

planning,explorationplanning,experimentation,etc.,somecombinationofgoal-basedandutility-basedagentswouldbeneeded.基于模型的映射智能体能够满足低水平的航线和避免障碍，为了路由计划，探测计划，实验等。

这需要基于目标和效用的智能体。

d.Forspecificprooftasks,agoal-basedagentisneeded.For“exploratory”tasks—e.g.,“Provesomeusefullemmataconcerningoperationsonstrings”—autility-basedarchitecturemightbeneeded.

为了明确的检验任务，需要基于目标的智能体，为探测任务，

任务环境

机器人足

可观察性

确定性

片段性

静态性

离散型

智能体数

球运动员

部分

随机的

连续的

动态的

连续的

多

因特网购

书智能体

部分

确定的

连续的

静态的

离散的

单

自主火星

漫步者

部分

随机的

连续的

动态的

连续的

单

数学家的定理

证明助手

完全

确定的

连续的

静态的

离散的

多

3.1Defineinyourownwordsthefollowingterms:

state,statespace,searchtree,searchnode,goal,action,successorfunction,andbranchingfactor.

•state:

Astateisasituationthatanagentcanfinditselfin.Wedistinguishtwotypesofstates:

worldstates（theactualconcretesituationsintherealworld）andrepresentationalstates（theabstractdescriptionsoftherealworldthatareusedbytheagentindeliberatingaboutwhattodo）.

•statespace:

Astatespaceisagraphwhosenodesarethesetofallstates,andwhoselinksareactionsthattransformonestateintoanother.

•searchtree:

Asearchtreeisatree（agraphwithnoundirectedloops）inwhichtherootnodeisthestartstateandthesetofchildrenforeachnodeconsistsofthestatesreachablebytakinganyaction.

•searchnode:

Asearchnodeisanodeinthesearchtree.

•goal:

Agoalisastatethattheagentistryingtoreach.

•action:

Anactionissomethingthattheagentcanchoosetodo.

•successorfunction:

Asuccessorfunctiondescribedtheagent’soptions:

givenastate,itreturnsasetof

（action,state）pairs,whereeachstateisthestatereachablebytakingtheaction.

•branchingfactor:

Thebranchingfactorinasearchtreeisthenumberofactionsavailabletotheagent.

3.7Givetheinitialstate,goaltest,successorfunction,andcostfunctionforeachofthefollowing.Chooseaformulationthatispreciseenoughtobeimplemented.

a.Youhavetocoloraplanarmapusingonlyfourcolors,insuchawaythatnotwoadjacentregionshavethesamecolor.

Initialstate:

Noregionscolored.

Goaltest:

Allregionscolored,andnotwoadjacentregionshavethesamecolor.Successorfunction:

Assignacolortoaregion.

Costfunction:

Numberofassignments.路径耗损

b.A3-foot-tallmonkeyisinaroomwheresomebananasaresuspendedfromthe8-footceiling.Hewouldliketogetthebananas.Theroomcontainstwostackable,movable,climbable3-foot-highcrates.

Initialstate:

Asdescribedinthetext.初始状态：

Goaltest:

Monkeyhasbananas.目标测试：

猴子拿到香蕉

后继函数:

Hoponcrate;Hopoffcrate;Pushcratefromonespottoanother;Walkfromonespottoanother;

grabbananas（ifstandingoncrate）.挪动箱子，，把箱子叠起，走到箱子上拿香蕉

Costfunction:

Numberofactions.行动数量

c.Youhaveaprogramthatoutputsthemessage“illegalinputrecord”whenfedacertainfileofinputrecords.Youknowthatprocessingofeachrecordisindependentoftheotherrecords.Youwanttodiscoverwhatrecordisillegal.

Initialstate:

consideringallinputrecords.

Goaltest:

consideringasinglerecord,anditgives“illegalinput”message.

Successorfunction:

runagainonthefirsthalfoftherecords;runagainonthesecondhalfoftherecords.

Costfunction:

Numberofruns.

Note:

Thisisacontingencyproblem;youneedtoseewhetherarungivesanerrormessageornottodecidewhattodonext.

d.Youhavethreejugs,measuring12gallons,8gallons,and3gallons,andawaterfaucet.Youcanfillthejugsuporemptythemoutfromonetoanotherorontotheground.Youneedtomeasureoutexactlyonegallon.

Initialstate:

jugshavevalues[0,0,0].

Successorfunction:

givenvalues[x,y,z],generate[12,y,z],[x,8,z],[x,y,3]（byfilling）;[0,y,z],[x,0,z],[x,y,0]（byemptying）;orforanytwojugswithcurrentvaluesxandy,pouryintox;thischangesthejugwithxtotheminimumofx+yandthecapacityofthejug,anddecrementsthejugwithybytheamountgainedbythefirstjug.

Costfunction:

Numberofactions.

3.8Consider