Year
2017
Page
1-15
Source
Computer Speech&Language
Format Published
PDF
Abstract
Weproposeamulti-stylelearning(multi-styletrainingCdeeplearning)procedurethatreliesondeepdenoisingautoencoders(DAEs)toextractandorganizethe mostdiscriminativeinformationinatraining database.Traditionally,multi-styletrainingpro-ceduresrequire eithercollectingorartificiallycreatingdatasamples (e.g.,bynoiseinjectionordatacombination)andtrainingadeepneuralnetwork (DNN)withallofthesedifferentconditions.Toexpandtheapplicabilityofdeeplearning, thepresentstudyinsteadadoptsaDAEtoaugmenttheoriginaltrainingset.First, aDAEisutilizedtosynthesizedatathatcapturesusefulstructureintheinputdistribution.Next, thissyntheticdataiscombinedandmixedwithintheoriginaltrainingsetto exploitthepowerfulcapabilitiesofDNNclassifierstolearnthe complexdecisionboundariesinheterogeneousconditions. ByassigningaDAEtosyn-thesizeadditionalexamplesofrepresentativevariations ,multi-stylelearningmakesclassboundarieslesssensitivetocorruptionsbyenforcingback -endDNNstoemphasizeonthemostdiscriminativepatterns.Moreover ,thisdeeplearningtechniquemitigatesthecostandtimeofdata collectionandiseasytoincorporateintotheinternetofthings (IoT).Resultsshowedthesedata-mixedDNNsprovidedconsistentperformanceimprovements withoutevenrequiringanypreprocessingonthetestsets.
Call. No.
EA 49
IndexDate
1397/10/15
Indexer
Dashagha
Title of Article
Multi-style learning with denoising auto encodersforacoustic modeling in the internet of things(IoT)
RecordNumber
50
Author/Authors