Optimizing FPGA Acceleration for Recurrent Neural Network Language Model