Tight Bounds on the Smallest Eigenvalue of the Neural Tangent Kernel for Deep ReLU Networks

TightBoundsontheSmallestEigenvalueoftheNeuralTangentKernelforDeepReLUNetworksQuynhNguyen1MarcoMondelli2GuidoMontufar13AbstractWeassumethatthenetworkhasasingleoutput,namelynL=1andWL∈RnL−1.Forconsi...