Training Large Neural Networks With Low-Dimensional Error Feedback

Maher Hanut, Jonathan Kadmon

Training deep neural networks typically relies on backpropagating high dimensional error signals a computationally intensive process with little evidence supporting its implementat…