Abstract:Learningfromafewexamplesremainsakeychallengeinmachinelearning.Despiterecentadvancesinimportantdomainssuchasvisionandlanguage,thestandardsuperviseddeeplearningparadigmdoesnotofferasatisfactorysolutionforlearningnewconceptsrapidlyfromlittledata.Inthiswork,weemployideasfrommet