In recent years significant progress has been made in successfully training
recurrent neural networks (RNNs) on sequence learning problems involving long
range temporal dependencies. The progress has been made on three fronts: (a)
Algorithmic improvements involving sophisticated optimi