Power-of-Two Quantization for Efficient FPGA-Based GRU Architectures