没有合适的资源?快使用搜索试试~ 我知道了~
EN-MakeYourOwnNeuralNetwork.pdf
需积分: 5 0 下载量 4 浏览量
2024-01-04
15:35:06
上传
评论
收藏 9.33MB PDF 举报
温馨提示
试读
224页
EN-MakeYourOwnNeuralNetwork
资源推荐
资源详情
资源评论
Contents
Prologue
TheSearchforIntelligentMachines
ANatureInspiredNewGoldenAge
Introduction
Whoisthisbookfor?
Whatwillwedo?
Howwillwedoit?
Author’sNote
Part1HowTheyWork
EasyforMe,HardforYou
ASimplePredictingMachine
ClassifyingisNotVeryDifferentfromPredicting
TrainingASimpleClassifier
SometimesOneClassifierIsNotEnough
Neurons,Nature’sComputingMachines
FollowingSignalsThroughANeuralNetwork
MatrixMultiplicationisUseful..Honest!
AThreeLayerExamplewithMatrixMultiplication
LearningWeightsFromMoreThanOneNode
BackpropagatingErrorsFromMoreOutputNodes
BackpropagatingErrorsToMoreLayers
BackpropagatingErrorswithMatrixMultiplication
HowDoWeActuallyUpdateWeights?
WeightUpdateWorkedExample
PreparingData
Part2DIYwithPython
Python
InteractivePython=IPython
AVeryGentleStartwithPython
NeuralNetworkwithPython
TheMNISTDatasetofHandwrittenNumbers
1
Part3EvenMoreFun
YourOwnHandwriting
InsidetheMindofaNeuralNetwork
CreatingNewTrainingData:Rotations
Epilogue
AppendixA:
AGentleIntroductiontoCalculus
AFlatLine
ASlopedStraightLine
ACurvedLine
CalculusByHand
CalculusNotByHand
CalculuswithoutPlottingGraphs
Patterns
FunctionsofFunctions
YoucandoCalculus!
AppendixB:
DoItwithaRaspberryPi
InstallingIPython
MakingSureThingsWork
TrainingAndTestingANeuralNetwork
RaspberryPiSuccess!
2
Prologue
TheSearchforIntelligentMachines
Forthousandsofyears,wehumanshavetriedtounderstandhowourownintelligenceworks
andreplicateitinsomekindofmachinethinkingmachines.
We’venotbeensatisfiedbymechanicalorelectronicmachineshelpinguswithsimpletasks
flintsparkingfires,pulleysliftingheavyrocks,andcalculatorsdoingarithmetic.
Instead,wewanttoautomatemorechallengingandcomplextaskslikegroupingsimilarphotos,
recognisingdiseasedcellsfromhealthyones,andevenputtingupadecentgameofchess.
Thesetasksseemtorequirehumanintelligence,oratleastamoremysteriousdeepercapability
ofthehumanmindnotfoundinsimplemachineslikecalculators.
Machineswiththishumanlikeintelligenceissuchaseductiveandpowerfulideathatourculture
isfulloffantasies,andfears,aboutittheimmenselycapablebutultimatelymenacingHAL
9000inStanleyKubrick’s2001:ASpaceOdyssey
,thecrazedactionTerminator
robotsandthe
talkingKITTcarwithacoolpersonalityfromtheclassicKnightRider
TVseries.
WhenGaryKasparov,thereigningworldchesschampionandgrandmaster,wasbeatenbythe
IBMDeepBluecomputerin1997wefearedthepotentialofmachineintelligencejustasmuch
aswecelebratedthathistoricachievement.
Sostrongisourdesireforintelligentmachinesthatsomehavefallenforthetemptationtocheat.
TheinfamousmechanicalTurkchessmachinewasmerelyahiddenpersoninsideacabinet!
3
ANatureInspiredNewGoldenAge
Optimismandambitionforartificialintelligencewereflyinghighwhenthesubjectwasformalised
inthe1950s.Initialsuccessessawcomputersplayingsimplegamesandprovingtheorems.
Somewereconvincedmachineswithhumanlevelintelligencewouldappearwithinadecadeor
so.
Butartificialintelligenceprovedhard,andprogressstalled.The1970ssawadevastating
academicchallengetotheambitionsforartificialintelligence,followedbyfundingcutsanda
lossofinterest.
Itseemedmachinesofcoldhardlogic,ofabsolute1sand0s,wouldneverbeabletoachieve
thenuancedorganic,sometimesfuzzy,thoughtprocessesofbiologicalbrains.
Afteraperiodofnotmuchprogressanincrediblypowerfulideaemergedtoliftthesearchfor
machineintelligenceoutofitsrut.Whynottrytobuildartificialbrainsbycopyinghowreal
biologicalbrainsworked?Realbrainswithneuronsinsteadoflogicgates,softermoreorganic
reasoninginsteadofthecoldhard,blackandwhite,absolutisttraditionalalgorithms.
Scientistwereinspiredbytheapparentsimplicityofabeeorpigeon'sbraincomparedtothe
complextaskstheycoulddo.Brainsafractionofagramseemedabletodothingslikesteer
flightandadapttowind,identifyfoodandpredators,andquicklydecidewhethertofightor
4
剩余223页未读,继续阅读
资源评论
死磕代码程序媛
- 粉丝: 108
- 资源: 316
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功