简介:Thecapabilityofmultilayerperceptrons(MLPs)forapproximatingcontinuousfunctionswitharbitraryaccuracyhasbeendemonstratedinthepastdecades.Backpropagation(BP)algorithmisthemostpopularlearningalgorithmfortrainingofMLPs.Inthispaper,asimpleiterationformulaisusedtoselecttheleamingrateforeachcycleoftrainingprocedure,andaconvergenceresultispresentedfortheBPalgo-rithmfortrainingMLPwithahiddenlayerandalinearoutputunit.Themonotonicityoftheerrorfunctionisalsoguaranteedduringthetrainingiteration.
简介:Letfibeaboundedsmoothdomain.Inthispaper,theauthorsdefinetheBesovspacesBpa,pon,establishtheatomicdecompositionofthesespaces,andobtaintheregularityestimateoftheDirichletproblemandtheNeumannproblemfortheLaplaceoperatoronthesespaces.