Journal of IKEEE (전기전자학회논문지)

Institute of Korean Electrical and Electronics Engineers (한국전기전자학회)
권호 표지
DOI QR코드

DOI QR Code

MLP accelerator implementation by approximation of activation function

활성화 함수의 근사화를 통한 MLP 가속기 구현

  • Received : 2018.03.23
  • Accepted : 2018.03.29
  • Published : 2018.03.31
Download PDF
⟨ Previous Next ⟩

Abstract

In this paper, sigmoid function, which is difficult to implement at hardware level and has a slow speed, is approximated by using PLAN. We use this as an activation function of MLP structure to reduce resource consumption and speed up. In this paper, we show that the proposed method maintains 95% accuracy in $5{\times}5$ size recognition and 1.83 times faster than GPGPU. We have found that even with similar resources as MLPA accelerators, we use more neurons and converge at higher accuracy and higher speed.

본 논문에서는 하드웨어레벨로 구현이 어렵고 속도가 느린 sigmoid 함수를 PLAN을 이용하여 근사치로 출력하였다. 이를 MLP 구조의 활성화 함수로 사용하여 자원소모를 줄이고 속도를 개선하고자 하였다. 본 논문에서 제안하는 방법은 $5{\times}5$크기의 숫자 인식에 약 95%의 정확도를 유지하면서 GPGPU보다 약 1.83배의 빠른 속도를 보였다. 또한 MLPA가속기와 비슷한 자원을 사용함에도 더 많은 뉴런을 사용하여 높은 정확도에 빠른 속도로 수렴하는 것을 확인하였다.

Keywords

References

  1. Nara Im, "System identification of plane frames using an artificial neural network," Donga University master degree, 2001.
  2. TaeHwan Kim "Investigations on representing latent space features for dynamic system control based on machine learning methods," Korea University master degree, 2017.
  3. Youngsu Kim, "The prediction of surface settlement and tunnel's behavior during tunnel excavation by using Artificial neural network," Kyungpook National University master degree , 2007.
  4. EuiSun Lee, "Research on Safety Estimation of Amusement Devices Structure by Artificial Neural Network," Konyang University master degree , 2011.
  5. Panca Mudji Rahardjo, Moch. Rif'an dan Nanang Sulistyanto, "The Implementation of Feedforward Backpropagation Algorithm for Digit handwritten Recognition in a Xilinx Spartan-3," Jurnal EECCIS, vol.4, no.2, pp17-21, 2010.