Optimization and compression methods for recurrent neural networks on-chip Relatore: