Recurrent Neural Network

A Recurrent Neural Network (RNN) are mainly used to process sequential data (text, speech, time series, …), where order of element is important. It has a backpropagation mechanism to feed the results back to the network.

RNNs have a better capacity to retain information over a sequence, Compared to CNNs which doesn't have explicit memory. It's also harder to parallelize, since each step of the sequence need to be computed before the next one.