10_14_recursive_network_training

10_14_recursive_network_training

ID:39713775

大小:3.54 MB

页数:7页

时间:2019-07-09

10_14_recursive_network_training_第1页
10_14_recursive_network_training_第2页
10_14_recursive_network_training_第3页
10_14_recursive_network_training_第4页
10_14_recursive_network_training_第5页
资源描述:

《10_14_recursive_network_training》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、NeuralnetworksNaturallanguageprocessing-recursivenetworktrainingtoformawhole.Weintroduceamax-margin2structurepredictionarchitecturebasedonre-cursiveneuralnetworksthatcansuccessfullyRECURSIVENEURALNETWORKrecoversuchstructurebothincomplexsceneimagesaswellassentences.Thesamealgo-

2、Topics:recursiveneuralnetwork(RNN)rithmcanbeusedbothtoprovideacompeti-•Idea:recursivelymergepairsofword/phraserepresentationstivesyntacticparserfornaturallanguagesen-tencesfromthePennTreebankandtoout-performalternativeapproachesforsemanticscenesegmentation,annotationandclassifi

3、-cation.Forsegmentationandannotationouralgorithmobtainsanewlevelofstate-of-the-wordrepresentations{artperformanceontheStanfordbackgrounddataset(78.1%).Thefeaturesfromtheim-ageparsetreeoutperformGistdescriptorsfor•Weneed2thingsFigure1.Illustrationofourrecursiveneuralnetworkar-S

4、ocher,Lin,NgandManning,2011sceneclassificationby4%.chitecturewhichparsesimagesandnaturallanguagesen-‣amodelthatmergespairsofrepresentationstences.Segmentfeaturesandwordindices(orange)are‣amodelthatdeterminesthetreestructurefirstmappedintosemanticfeaturespace(blue)andthen1.Introd

5、uctionrecursivelymergedbythesameneuralnetworkuntiltheyrepresenttheentireimageorsentence.BothmappingsandRecursivestructureiscommonlyfoundindifferentmergingsarelearned.modalities,asshowninFig.1.Thesyntacticrulescanberecursivelysplitintosmallercarregionsdepict-ofnaturallanguageare

6、knowntoberecursive,withingpartssuchastiresandwindowsandthesepartscannounphrasescontainingrelativeclausesthatthem-occurinothercontextssuchasbeneathairplanesorinselvescontainnounphrases,e.g.,...thechurchwhichhouses.Weshowthatrecoveringthisstructurehelpshasnicewindows....Similarl

7、y,onefindsnestedhier-inunderstandingandclassifyingsceneimages.Inthisarchicalstructuringinsceneimagesthatcapturebothpaper,weintroducerecursiveneuralnetworks(RNNs)part-ofandproximityrelationships.Forinstance,carsforpredictingrecursivestructureinmultiplemodali-areoftenontopofstree

8、tregions.Alargecarregionties.Weprimarilyfocusonsceneunderstan

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。
相关文章
更多
相关标签