@INCOLLECTION{MR1114761,
   AUTHOR       = {E.D. Sontag},
   BOOKTITLE    = {Neural networks (New Brunswick, NJ, 1990)},
   PUBLISHER    = {Academic Press},
   TITLE        = {Capabilities and training of feedforward nets},
   YEAR         = {1991},
   ADDRESS      = {Boston, MA},
   OPTCHAPTER   = {},
   OPTCROSSREF  = {},
   OPTEDITION   = {},
   OPTEDITOR    = {},
   OPTMONTH     = {},
   OPTNOTE      = {},
   OPTNUMBER    = {},
   PAGES        = {303--321},
   OPTSERIES    = {},
   OPTTYPE      = {},
   OPTVOLUME    = {},
   KEYWORDS     = {neural networks, neural networks},
   PDF          = {../../FTPDIR/90caip.pdf},
   ABSTRACT     = { This paper surveys recent work by the author on 
      learning and representational capabilities of feedforward nets. The 
      learning results show that, among two possible variants of the 
      so-called backpropagation training method for sigmoidal nets, both of 
      which variants are used in practice, one is a better generalization 
      of the older perceptron training algorithm than the other. The 
      representation results show that nets consisting of sigmoidal neurons 
      have at least twice the representational capabilities of nets that 
      use classical threshold neurons, at least when this increase is 
      quantified in terms of classification power. On the other hand, 
      threshold nets are shown to be more useful when approximating 
      implicit functions, as illustrated with an application to a typical 
      control problem. }
}

{"_id":"wB9x3Jqyo2pETGgm8","bibbaseid":"sontag-capabilitiesandtrainingoffeedforwardnets-1991","downloads":0,"creationDate":"2018-10-18T05:07:06.590Z","title":"Capabilities and training of feedforward nets","author_short":["Sontag, E."],"year":1991,"bibtype":"incollection","biburl":"http://www.sontaglab.org/PUBDIR/Biblio/complete-bibliography.bib","bibdata":{"bibtype":"incollection","type":"incollection","author":[{"firstnames":["E.D."],"propositions":[],"lastnames":["Sontag"],"suffixes":[]}],"booktitle":"Neural networks (New Brunswick, NJ, 1990)","publisher":"Academic Press","title":"Capabilities and training of feedforward nets","year":"1991","address":"Boston, MA","optchapter":"","optcrossref":"","optedition":"","opteditor":"","optmonth":"","optnote":"","optnumber":"","pages":"303–321","optseries":"","opttype":"","optvolume":"","keywords":"neural networks, neural networks","pdf":"../../FTPDIR/90caip.pdf","abstract":"This paper surveys recent work by the author on learning and representational capabilities of feedforward nets. The learning results show that, among two possible variants of the so-called backpropagation training method for sigmoidal nets, both of which variants are used in practice, one is a better generalization of the older perceptron training algorithm than the other. The representation results show that nets consisting of sigmoidal neurons have at least twice the representational capabilities of nets that use classical threshold neurons, at least when this increase is quantified in terms of classification power. On the other hand, threshold nets are shown to be more useful when approximating implicit functions, as illustrated with an application to a typical control problem. ","bibtex":"@INCOLLECTION{MR1114761,\n AUTHOR = {E.D. Sontag},\n BOOKTITLE = {Neural networks (New Brunswick, NJ, 1990)},\n PUBLISHER = {Academic Press},\n TITLE = {Capabilities and training of feedforward nets},\n YEAR = {1991},\n ADDRESS = {Boston, MA},\n OPTCHAPTER = {},\n OPTCROSSREF = {},\n OPTEDITION = {},\n OPTEDITOR = {},\n OPTMONTH = {},\n OPTNOTE = {},\n OPTNUMBER = {},\n PAGES = {303--321},\n OPTSERIES = {},\n OPTTYPE = {},\n OPTVOLUME = {},\n KEYWORDS = {neural networks, neural networks},\n PDF = {../../FTPDIR/90caip.pdf},\n ABSTRACT = { This paper surveys recent work by the author on \n learning and representational capabilities of feedforward nets. The \n learning results show that, among two possible variants of the \n so-called backpropagation training method for sigmoidal nets, both of \n which variants are used in practice, one is a better generalization \n of the older perceptron training algorithm than the other. The \n representation results show that nets consisting of sigmoidal neurons \n have at least twice the representational capabilities of nets that \n use classical threshold neurons, at least when this increase is \n quantified in terms of classification power. On the other hand, \n threshold nets are shown to be more useful when approximating \n implicit functions, as illustrated with an application to a typical \n control problem. }\n}\n\n","author_short":["Sontag, E."],"key":"MR1114761","id":"MR1114761","bibbaseid":"sontag-capabilitiesandtrainingoffeedforwardnets-1991","role":"author","urls":{},"keyword":["neural networks","neural networks"],"downloads":0,"html":""},"search_terms":["capabilities","training","feedforward","nets","sontag"],"keywords":["neural networks","neural networks"],"authorIDs":["5bc814f9db768e100000015a"],"dataSources":["DKqZbTmd7peqE4THw"]}