Model Compression - Cornell University

ModelCompressionCristianBucil hundredsor thousandsof base-level classi , thespacerequiredto storethismany clas-si ers,andthetimerequiredto executethemat run-time,prohibitstheirusein applicationswheretestsetsarelarge( ),wherestoragespaceis ata premium( ),andwherecomputationalpower is limited( ).We present a method for\compressing"large,complexensemblesin to smaller,fastermodels,usuallywith-outsign i cant lossin Subject [PatternRe-cognition]:Models{ :Algorithms,Experimentation,Measure-ment , Performance, :SupervisedLearning, a collectionof modelswhosepredictionsarecombinedby weightedaveragingor beenthefocusof signi cant research in thepastdecade,anda variety of ensemblemethods have knownensemblemethods includebagging[2],boosting[14],randomfor ests[3],Bayesianavera}

model such as a neural net with little loss in performance. An important question is how do we get the pseudo data. In some domains large amounts of unlabeled data is easy to collect (e.g. in text, web and image domains) and can be used as pseudo data. In other domains, however, unla-beled data is not readily available and synthetic cases need

Loading..

Tags:

  Model, Data, Compression, Model compression

Information

Related search queries