Gravar-mail: Learning Without Feedback: Fixed Random Learning Signals Allow for Feedforward Training of Deep Neural Networks