Direct Feedback Alignment Provides Learning in Deep Neural Networks

Arild Nøkland

Artificial neural networks are most commonly trained with the back-propagation algorithm, where the gradient for learning is provided by back-propagating the error, layer by layer,…