Figure 8: LSTM optimization of a recurrent neural network unit.