Scaled and Graduated Learning in Deep RELU Networks and Reconstructing Depp Inelastic Scattering Kinematics