Measurements of Three-Level Hierarchical Structure in the Outliers in the Spectrum of Deepnet Hessians

MeasurementsofThree-LevelHierarchicalStructureintheOutliersintheSpectrumofDeepnetHessiansVardanPapyan1Abstract(f(xi,c;θ),yc)∈R+isthecross-entropylossbetweenthesoftmaxoff(xi,c;θ)andtheone-hotvect...