Backpropagation

A method for training DNNs. Involves a forward pass (input to output prediction) and a backward pass (calculating error using a loss/cost function and propagating it back through layers to assign responsibility to weights for adjustment).