neural networks and pattern recognition

neural networks and pattern recognition

ID:14863416

大小:20.92 MB

页数:359页

时间:2018-07-30

neural networks and pattern recognition_第1页
neural networks and pattern recognition_第2页
neural networks and pattern recognition_第3页
neural networks and pattern recognition_第4页
neural networks and pattern recognition_第5页
资源描述:

《neural networks and pattern recognition》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库

1、PrefaceThisvolumerepresentsaturningpointinneuralnetworkadvancements.Thefirstneuralnetworksposed,suchasthemultilayerperceptron,werestaticnetworksthatclassifiedstaticpatterns—fixedvectors—andresultedinanetworkoutputthatwasyetanotherstaticpattern,anotherfix

2、ed-valuedvector.Neitherpatternchangedwithtime.Todaythefieldofneuralnetworksisadvancingbeyondthesestaticneu•ralnetworks,tomoreadvancedconceptsthatincorporatetime-dynamicsintheirinputs,outputs,andinternalprocessing.Neuralnetworksnowcanaccept,asinput,time-v

3、aryingsignals,evenmultichannelsignalsthatcor•respondtoavectororimagethatchangesovertime,andoftenprovideclassificationofdatathatvariesovertime.Somenetworksproduceresultsthataretime-dynamic,includingoscillationsandtemporalpatterns,andsometimesself-sustaine

4、dactivitycanbeasignatureuniquetothenetwork'sstructureortothepatternsthatstimulatethenetwork.Whataretheelementsandarchitecturesthatmakeitpossibletoad•vancefromstaticarchitecturestodynamiccomputation?Whatapproachesprovideincreasedcapabilitiesforneuralnetwo

5、rks?Thesequestionsarean•swered,inpart,bythisvolume.Pulse-coupledneuralnetworksincorporateprocessingelements,neurons,thatcommunicatebysendingpulsestooneanother.Pulse-coupledneuralnetworkscanrepresentspatialinformationinthetimestructureoftheiroutputpulsetr

6、ainsandcansegmentanimageintomulti-neurontime-synchronousgroups.Johnson,Ranganath,Kuntimad,andCaulfield,inChapter1,illustratethesecapabilitiesandshowthearchitecturalstruc•tureofthepulse-couplednetworks.Motionperceptionisanessentialcapabilityforadvancedorg

7、anisms,yettheabilitytodetectmotionsandimageflowcomputationallyisadifficultproblem.InChapter2,LiandWangproposearecurrentneuralnetworkmodelthatcanbeoperatedasynchronouslyinparalleltoachieveareal•timesolution.InChapter3,temporalpatternmatchingisperformedwhe

8、ndynamictimewarpingiscombinedwithaHopfieldnetwork.UnalandTepedelenli-oglushowhowadynamicprogrammingalgorithmthatcomparesaninputtestsignalwithareferencetemplatesignal,reducingthenonlineartimemisalignmentsbetweenthetwopatter

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。