QuantifyingtheBenefitofUsingDifferentiableLearningoverTangentKernelsEranMalach1PritishKamath2EmmanuelAbbe3NathanSrebro2AbstractCanallthesuccessofdeeplearningbeexplainedusingtheNTK?Thiswouldimplytha...
ProvablyStrictGeneralisationBenefitforEquivariantModelsBrynElesedy1SheheryarZaidi2AbstractasSokolicetal.(2017);Sannai&Imaizumi(2019),coveronlytheworstcaseperformanceofalgorithms.TheseworksItiswidel...
OntheGeneralizationBenefitofNoiseinStochasticGradientDescentSamuelL.Smith1ErichElsen1SohamDe1Abstractbatches,orbecauseauthorsoftencomparedifferentbatchsizesunderaconstantepochbudget(suchthatsmallba...