On the Random Conjugate Kernel and Neural Tangent Kernel

                        Zhengmian Hu 1 Heng Huang 1

              Abstract               due to the difficulties raised by the non-convexity of the loss
                                 function and the complication of optimization methods.
    We investigate the distributions of Conjugate Ker-
    nel (CK) and Neural Tangent Kernel (NTK) for        One way to simplify the DNNs analysis is to only train
    ReLU networks with random init ...