DOI QR코드

DOI QR Code

IoT 컴퓨팅 환경을 위한 뉴로모픽 기반 플랫폼의 추론시간 단축

Reduction of Inference time in Neuromorphic Based Platform for IoT Computing Environments

  • 김재섭 (숭실대학교 일반대학원 컴퓨터학과) ;
  • 이승연 (숭실대학교 일반대학원 컴퓨터학과) ;
  • 홍지만 (숭실대학교 컴퓨터학부)
  • 투고 : 2022.03.08
  • 심사 : 2022.03.24
  • 발행 : 2022.03.31

초록

뉴로모픽 아키텍처는 스파이킹 신경망(SNN, Spiking Neural Network) 모델을 사용하여, 추론 실험을 통해 스파이크 값이 많이 누적될수록 정확한 결과를 도출한다. 추론 결과가 특정 값으로 수렴할 경우, 추론 실험을 더 진행해도 결과의 변화가 작아 소비 전력이 더 커질 수 있다. 특히, 인공지능 기반 IoT 환경에서는 전력 낭비는 큰 문제가 될 수 있다. 따라서 본 논문에서는 뉴로모픽 아키텍처 환경에서 추론 이미지 노출 시간을 조절하여 추론 시간을 단축함으로써 인공지능 기반 IoT의 전력 낭비를 줄이는 기법을 제안한다. 제안한 기법은 추론 정확도의 변화를 반영하여 다음 추론 이미지 노출 시간을 계산한다. 또한, 추론 정확도의 변화량 반영비율을 계수 값으로 조절할 수 있으며, 다양한 계수 값의 비교 실험을 통해 최적의 계수 값을 찾는다. 제안한 기법은 목표 정확도에 해당하는 추론 이미지 노출 시간은 선형 기법보다 크지만 최종 추론 시간은 선형 기법보다 적다. 제안한 기법의 성능을 측정하고 평가한 결과, 제안한 기법을 적용한 추론 실험이 선형 기법을 적용한 추론 실험보다 최종 노출 시간을 약 90% 단축할 수 있음을 확인한다.

The neuromorphic architecture uses a spiking neural network (SNN) model to derive more accurate results as more spike values are accumulated through inference experiments. When the inference result converges to a specific value, even if the inference experiment is further performed, the change in the result is smaller and power consumption may increase. In particular, in an AI-based IoT environment, power consumption can be a big problem. Therefore, in this paper, we propose a technique to reduce the power consumption of AI-based IoT by reducing the inference time by adjusting the inference image exposure time in the neuromorphic architecture environment. The proposed technique calculates the next inferred image exposure time by reflecting the change in inference accuracy. In addition, the rate of reflection of the change in inference accuracy can be adjusted with a coefficient value, and an optimal coefficient value is found through a comparison experiment of various coefficient values. In the proposed technique, the inference image exposure time corresponding to the target accuracy is greater than that of the linear technique, but the overall power consumption is less than that of the linear technique. As a result of measuring and evaluating the performance of the proposed method, it is confirmed that the inference experiment applying the proposed method can reduce the final exposure time by about 90% compared to the inference experiment applying the linear method.

키워드

과제정보

이 논문은 2016년도 정부(교육부)의 재원으로 한국연구재단의 지원을 받아 수행된 연구임 (No. NRF-2016R1D1A1B01016073, 가상화 기반 오픈 게이트웨이 플랫폼과 오케스트레이션 보안 서비스 프레임워크 설계 및 구현).

참고문헌

  1. Younggwan Kim, Jusuk Lee, Ajung Kim, Jiman Hong, "Load Balancing Scheme for Machine Learning Distributed Environment," Smart Media Journal, Vol.10, No.1, pp. 25-31, 2021.
  2. Sun Park, Jongwon Kim, "Red Tide Algea Image Classification using Deep Learning based Open Source," Smart Media Journal, Vol. 7, No. 2, pp34-39, 2018. https://doi.org/10.30693/SMJ.2018.7.2.34
  3. Seo jeong Kim, Jae Su Lee, Hyong Suk Kim, "Deep learning-based Automatic Weed Detection on Onion Field," Smart Media Journal, Vol. 7, No.3, pp16-21, 2018. https://doi.org/10.30693/SMJ.2018.7.3.16
  4. Hayun Lee, Dongkun Shin, "Performance and Energy Comparison of Different BLAS and Neural Network Libraries for Efficient Deep Learning Inference of ARM-baased IoT Devices," Journal of KIISE, Vol. 46, No. 3, pp. 219-227, Mar. 2019. https://doi.org/10.5626/jok.2019.46.3.219
  5. Boseon Hong, Bongjae Kim, "Hardware Technology Trends for Neuromorphic Computing," Communications of the Korean Institute of Information Scientists and Engineers, Vol 38, No 2, pp. 32-39, Feb. 2020.
  6. Seoyeon Kim, Young-Sun Yun, Jinman Jung, "Design of a Framework for supporting Neuromorphic Hardware in IoT platform," Communications of the Korean Institute of Information Scientists and Engineers, Vol 38, No 2, pp. 51-57, Feb. 2020.
  7. Keon Myung Lee, "Behavior of Spiking Neuron Models and Learning of Spiking Neural Networks," Communications of the Korean Institute of Information Scientists and Engineers, Vol 38, No 2, pp. 8-19, Feb. 2020.
  8. Jangsaeng Kim, Dongseok Kwon, Sung Yun Woo, Won-Mook Kang, Soochang Lee, Seongbin Oh, Chul-Heung Kim, Jong-Ho Bae, Byung-Gook Park, and Jong-Ho Lee, "Hardware-based spiking neural network architecture using simplified backpropagation algorithm and homeostasis functionality," Neurocomputing, Vol. 428, pp. 153-165, Mar. 2021. https://doi.org/10.1016/j.neucom.2020.11.016
  9. Trevor Bekolay, James Bergstra, Eric Hunsberger, Travis DeWolf, Terrence Stewart, Daniel Rasmussen, Xuan Choo, Aaron Voelker, and Chris Eliasmith, "Nengo: a Python tool for building large-scale functional brain models," Frontiers in Neuroinformatics, Vol 7, No 48, pp. 1-13, Jan. 2014.
  10. Terrence C Stewart, "A technical overview of the neural engineering framework," Centre for Theoretical Neuroscience technical report, University of Waterloo, Canada, Vol 110, 2012.
  11. Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, vol. 86, no. 11, 2278-2323, Nov. 1998. https://doi.org/10.1109/5.726791
  12. LattePanda Alpha 864s(2021). https://www.lattepanda.com/products/lattepanda-alpha-864s.html (accessed Mar., 02, 2022).
  13. NengoLoihi(2022). https://www.nengo.ai/nengo-loihi/v1.1.0/index.html (accessed Feb., 25, 2022).