Log-quantization on GRU networks

Park, Sang-Ki; Park, Sang-Soo; Chung, Ki Seok

doi:10.1145/3290420.3290443

상세 보기

Log-quantization on GRU networks

Park, Sang-Ki ;
Park, Sang-Soo;
Chung, Ki Seok

Citations

SCOPUS

0

초록

Today, recurrent neural network (RNN) is used in various applications like image captioning, speech recognition and machine translation. However, because of data dependencies, recurrent neural network is hard to parallelize. Furthermore, to increase network’s accuracy, recurrent neural network uses complicated cell units such as long short-term memory (LSTM) and gated recurrent unit (GRU). To run such models on an embedded system, the size of the network model and the amount of computation need to be reduced to achieve low power consumption and low required memory bandwidth. In this paper, implementation of RNN based on GRU with a logarithmic quantization method is proposed. The proposed implementation is synthesized using high-level synthesis (HLS) targeting Xilinx ZCU102 FPGA running at 100MHz. The proposed implementation with an 8-bit log-quantization achieves 90.57% accuracy without re-training or fine-tuning. And the memory usage is 31% lower than that for an implementation with 32-bit floating point data representation.

키워드

AI; CNN; FPGA; HLS; HW/SW Co-Design; LeNet-5; SDSoC; Artificial intelligence; Digital arithmetic; Field programmable gate arrays (FPGA); Hardware-software codesign; High level synthesis; Logic Synthesis; Low power electronics; Speech recognition; Speech transmission; Data dependencies; Floating-point data; HW/SW Codesign; LeNet-5; Low-power consumption; Machine translations; Recurrent neural network (RNN); SDSoC; Long short-term memory

제목: Log-quantization on GRU networks

저자: Park, Sang-Ki ; Park, Sang-Soo; Chung, Ki Seok

DOI: 10.1145/3290420.3290443

발행일: 2018-11

유형: Conference Paper

저널명: ACM International Conference Proceeding Series

페이지: 112 ~ 116