認(rèn)知計(jì)算-helianghua1課件

認(rèn)知計(jì)算概述何良華學(xué)習(xí)目標(biāo)認(rèn)知概述認(rèn)知計(jì)算概述認(rèn)知計(jì)算經(jīng)典算法學(xué)習(xí)目標(biāo)認(rèn)知概述認(rèn)知計(jì)算概述認(rèn)知計(jì)算經(jīng)典算法1．思維

人腦對客觀現(xiàn)實(shí)間接的、概括的反應(yīng)，是認(rèn)識事物本質(zhì)特征及內(nèi)部規(guī)律的理性認(rèn)知過程。思維活動是人類認(rèn)識活動的最高形式，常通過語言文字表達(dá)思維具有連續(xù)性，否則為思維障礙。

抽象思維、洞察力和判斷力是反映思維水平的主要指標(biāo)。2．語言

是人們進(jìn)行思維的工具，是思維的物質(zhì)外殼學(xué)習(xí)語言的技巧與環(huán)境有關(guān)分接受性語言和表達(dá)性語言3．定向

人們對現(xiàn)實(shí)的感覺，對過去、現(xiàn)在、將來的察覺以及對自我存在的意識。包括時(shí)間定向、地點(diǎn)定向、空間定向和人物定向是大腦功能活動的綜合表現(xiàn)。即對環(huán)境的知覺狀態(tài)。4．意識

記憶-個(gè)人所經(jīng)歷過的事物在人腦中的反映，是人腦積累經(jīng)驗(yàn)的功能表現(xiàn)。評估方法-短時(shí)記憶-長時(shí)記憶（一）思維能力的評估－抽象思維概念力人腦反映客觀事物本質(zhì)特性的思維形式。通過抽象概括，把握事物的本質(zhì)特性而形成。評估方法通過數(shù)次健康教育后，請被評估者概括相關(guān)內(nèi)容（一）思維能力的評估－抽象思維推理力有已知判斷推出新判斷的思維過程歸納（從特殊到一般）演繹（從一般到特殊）評估方法根據(jù)被評估者的年齡特征提出一定的問題（一）思維能力的評估－抽象思維識別與理解客觀事物真實(shí)性的能力評估方法讓被評估者描述所處情形，再與實(shí)際情形作比較看有無差異你認(rèn)為導(dǎo)致你來就診的主要問題是什么？你如何判斷你目前的這種情況？（一）思維能力的評估－洞察力

肯定或否定某事物具有某種屬性或某行動方案可行性的思維方式受個(gè)體的年齡、情緒、智力、受教育水平、社會經(jīng)濟(jì)狀況、文化背景等的影響評估方法展示實(shí)物讓被評估者說出其屬性評價(jià)被評估者對將來打算的現(xiàn)實(shí)性與可行性進(jìn)行評估（一）思維能力的評估－判斷力

（二）語言能力的評估

語言能力是人們認(rèn)知水平的重要標(biāo)志，對判斷個(gè)體認(rèn)知水平很有價(jià)值。（二）語言能力的評估

評估方法

-提問-復(fù)述-自發(fā)性語言-命名-閱讀-書寫意識的臨床表現(xiàn)（四）意識的評估

影響認(rèn)知的因素年齡受教育水平生活經(jīng)歷文化背景疾病藥物作用酗酒吸毒

七個(gè)問題1.

認(rèn)識的本質(zhì)——兩條認(rèn)識路線的對立2.

認(rèn)識的能力3.

認(rèn)識的來源4.

認(rèn)識的過程5.

認(rèn)識的途徑6.

認(rèn)識的結(jié)果及其檢驗(yàn)7.

認(rèn)識的目的

認(rèn)識論

Epistemology

TheTheoryofKnowledge

哲學(xué)認(rèn)識論認(rèn)知

科學(xué)

——

無

所

不

能

——

學(xué)科認(rèn)知

科學(xué)

——

最

colorful

的

學(xué)科頭腦

風(fēng)暴

——

最

大

科學(xué)

發(fā)

現(xiàn)難

在

何處

——

最大

黑

箱黑

箱

方

法黑箱示意圖輸入輸出內(nèi)部機(jī)制已知已知未知人腦—黑箱變化已知對比推測伽

德

納

六

邊

形學(xué)科六邊形artificialintelligencen.人工智能Anthropologyn.人類學(xué)Linguisticsn.語言學(xué)Psychologyn.心理學(xué)Philosophyn.哲學(xué)Neurosciencen.神經(jīng)系統(tǒng)科學(xué)(指神經(jīng)病學(xué)、神經(jīng)化學(xué)等)學(xué)習(xí)目標(biāo)認(rèn)知概述認(rèn)知計(jì)算概述認(rèn)知計(jì)算經(jīng)典算法ComparisonofSiliconComputersandCarbonComputersDigitalcomputersare

MadefromsiliconAccurate(essentiallynoerrors)Fast(nanoseconds)Executelongchainsofseriallogical

operations(billions)IrritatingtohumansComparisonofSiliconComputersandCarbonComputersBrainsareMadefromcarboncompounds

Inaccurate(lowprecision,noisy)Slow(milliseconds,106timesslower)Executeshortchainsofparallelalogical

associativeoperations(perhaps10operations)UnderstandabletohumansPerformanceofSiliconComputersandCarbonComputerHugedisadvantageforcarbon:morethan1012

intheproductofspeedandpower.Butwedobetterandfasterthantheminmanytasks:speechrecognition,objectrecognition,facerecognition,motorcontrolmostcomplexmemoryfunctions,informationintegration.Implication:Cognitive“software”usesonlyafewbutverypowerfulelementaryoperations.WhyBuildaBrain-LikeComputer?

1.Engineering.

Computersareallspecialpurposedevices.

Manyoftheimportantpracticalcomputerapplicationsofthenextfewdecadeswillbecognitive:

·

Languageunderstanding.

·

Internetsearch.·

Cognitivedatamining.·

Decenthuman-computerinterfaces.

Wefeelitwillbenecessarytohaveabrain-likearchitecturetoruntheseapplicationsefficiently.2.KinshipRecognition,HumanFactors:

Toberecognizedasintelligentbyhumans,amachinehastohaveasomewhathuman-likeintelligence.Theremaybemanykindsofintelligence,butwecanonlyunderstandandcommunicatewithoneofthem!Successfulhuman-computerinteractionswillrequireabrain-likecomputerdoingcognitivecomputation.“Ifoxenandhorseshadhandsandcouldcreateworksofart,horseswoulddrawpicturesofgodslikehorsesandoxen,godslikeoxen…”Xenophanes(C.530B.C.E.)3.Personal:

Itwouldbetheultimatecoolgadget.Atechnologicalvision:In2050thepersonalcomputeryoubuyinWal-MartwillhavetwoCPU’swithverydifferentarchitecture:

First,atraditionalvonNeumannmachinethatrunsspreadsheets,doeswordprocessing,keepsyourcalendarstraight,etc.Whattheydonow.

Second,abrain-likechip

·

TohandletheinterfacewiththevonNeumannmachine,·

GiveyouthedatathatyouneedfromtheWeboryourfiles(butdidn’tthinktoaskfor).·

Beyoursiliconfriend,guide,andconfidant.History:TechnicalIssuesManyhaveproposedtheconstructionofbrain-likecomputersforcognitivecomputation.

Theseattemptsusuallystartwith

·

massivelyparallelarraysofneuralcomputingelements·

elementsbasedtosomedegreeonbiologicalneurons,·

thelayered2-Danatomyofmammaliancerebralcortex.

Suchattemptshavefailedcommercially.TheearlyconnectionmachinesfromThinkingMachines,Inc.,(W.D.Hillis,TheConnectionMachine,1987)wasthemostnearlysuccessfulcommercially..

Considertheextremesofcomputationalbrainmodels:FirstExtreme:BiologicalRealismThehumanbrainiscomposedofontheorderof1010neurons,connectedtogetherwithatleast1014neuralconnections.(Probablyunderestimates.)Biologicalneuronsandtheirconnectionsareextremelycomplexelectrochemicalstructures.Themorerealistictheneuronapproximationthesmallerthenetworkthatcanbemodeled.Thereisverygoodevidencethatforcerebralcortexabiggerbrainisabetterbrain.

Projectsthatmodelneuronsareofscientificinterest.

Theyarenotlargeenoughtomodelorsimulateinterestingcognition.

NeuralNetworks.

Themostsuccessfulbraininspiredmodelsareneuralnetworks.

Theyarebuiltfromsimpleapproximationsofbiologicalneurons:nonlinearintegrationofmanyweightedinputs.

Throwoutalltheotherbiologicaldetail.Cognitivecomputationisbasedonusefulapproximations.

SecondExtreme:AssociativelyLinkedNetworks.

Thesecondclassofbrain-likecomputingapproximationsisabasicpartofcomputerscience:

Associativelylinkedstructures.

Oneexampleofsuchastructureisasemanticnetwork.Suchstructuresunderliemostofthepracticallysuccessfulapplicationsofartificialintelligence.AssociativelyLinkedNetworks

(2)Theconnectionbetweenthebiologicalnervoussystemandsuchastructureisunclear.

Fewbelievethatnodesinasemanticnetworkcorrespondtosingleneuronsorgroupsofneurons.

Nodesarecomposedofmanypartsandcontainsignificantinternalstructure.

Physiology(fMRI)showsthatacomplexcognitivestructure–aword,forinstance–givesrisetowidelydistributedcorticalactivation.VirtueofLinkedNetworks:Theyhavesparselyconnectednodes.

Inpracticalsystems,thenumberoflinksconvergingonanoderangefromoneortwouptoadozenorso.

LookatSomeExamplesThebrain(andcognitivecomputation)dothingsdifferently:Ifyoubuildabrainexpecttogetweaknessesaswellasstrengths.Bothstrengthsandweaknessesareintrinsictothehardwareitself.Giveafewexamples.CognitiveStrengths

Strengths:Abilitytoapproximatecomplexeventsinusefulways(usingwords,concepts).Abilitytointegrateinformationfrommanysources.Effectivesearchofalargememory,thatis,integrationofpastexperiencewiththepresentsituation.Tightcouplingofhigher-levelcognitionwithperceptionNon-logicalprocessessuchas“intuition”forpredictionandunderstanding.CognitiveWeaknessesWeaknesses:

Higherrorrate.Slowresponsescomparedtosilicontimescales.Alogicalinformationprocessing,forexample,association.Oneresult:Greatdifficultywithlogicandformalreasoning.Lossofdetailinmemorystorage.Interferencefromothermemories.Prejudice(jumpingtoconclusions).Lackofexplanationforactions.ConclusionsBrainsareverydif