Efficient recurrent neural networks using structured matrices in FPGAS

Zhe Li, Shuo Wang, Caiwen Ding, Qinru Qiu, Yanzhi Wang, Yun Liang

Research output: Contribution to conferencePaper

Abstract

Recurrent Neural Networks (RNNs) are becoming increasingly important for time series-related applications which require efficient and real-time implementations. The recent pruning based work ESE (Han et al., 2017) suffers from degradation of performance/energy efficiency due to the irregular network structure after pruning. We propose block-circulant matrices for weight matrix representation in RNNs, thereby achieving simultaneous model compression and acceleration. We aim to implement RNNs in FPGA with highest performance and energy efficiency, with certain accuracy requirement (negligible accuracy degradation). Experimental results on actual FPGA deployments shows that the proposed framework achieves a maximum energy efficiency improvement of 35.7× compared with ESE.

Original languageEnglish (US)
StatePublished - Jan 1 2018
Event6th International Conference on Learning Representations, ICLR 2018 - Vancouver, Canada
Duration: Apr 30 2018May 3 2018

Conference

Conference6th International Conference on Learning Representations, ICLR 2018
CountryCanada
CityVancouver
Period4/30/185/3/18

Fingerprint

Recurrent neural networks
neural network
Energy efficiency
energy
efficiency
Field programmable gate arrays (FPGA)
Degradation
performance
time series
Time series
Recurrent Neural Networks
Energy

ASJC Scopus subject areas

  • Education
  • Computer Science Applications
  • Linguistics and Language
  • Language and Linguistics

Cite this

Li, Z., Wang, S., Ding, C., Qiu, Q., Wang, Y., & Liang, Y. (2018). Efficient recurrent neural networks using structured matrices in FPGAS. Paper presented at 6th International Conference on Learning Representations, ICLR 2018, Vancouver, Canada.

Efficient recurrent neural networks using structured matrices in FPGAS. / Li, Zhe; Wang, Shuo; Ding, Caiwen; Qiu, Qinru; Wang, Yanzhi; Liang, Yun.

2018. Paper presented at 6th International Conference on Learning Representations, ICLR 2018, Vancouver, Canada.

Research output: Contribution to conferencePaper

Li, Z, Wang, S, Ding, C, Qiu, Q, Wang, Y & Liang, Y 2018, 'Efficient recurrent neural networks using structured matrices in FPGAS', Paper presented at 6th International Conference on Learning Representations, ICLR 2018, Vancouver, Canada, 4/30/18 - 5/3/18.
Li Z, Wang S, Ding C, Qiu Q, Wang Y, Liang Y. Efficient recurrent neural networks using structured matrices in FPGAS. 2018. Paper presented at 6th International Conference on Learning Representations, ICLR 2018, Vancouver, Canada.
Li, Zhe ; Wang, Shuo ; Ding, Caiwen ; Qiu, Qinru ; Wang, Yanzhi ; Liang, Yun. / Efficient recurrent neural networks using structured matrices in FPGAS. Paper presented at 6th International Conference on Learning Representations, ICLR 2018, Vancouver, Canada.
@conference{2a782eeadca74e558c56ed044203ad9a,
title = "Efficient recurrent neural networks using structured matrices in FPGAS",
abstract = "Recurrent Neural Networks (RNNs) are becoming increasingly important for time series-related applications which require efficient and real-time implementations. The recent pruning based work ESE (Han et al., 2017) suffers from degradation of performance/energy efficiency due to the irregular network structure after pruning. We propose block-circulant matrices for weight matrix representation in RNNs, thereby achieving simultaneous model compression and acceleration. We aim to implement RNNs in FPGA with highest performance and energy efficiency, with certain accuracy requirement (negligible accuracy degradation). Experimental results on actual FPGA deployments shows that the proposed framework achieves a maximum energy efficiency improvement of 35.7× compared with ESE.",
author = "Zhe Li and Shuo Wang and Caiwen Ding and Qinru Qiu and Yanzhi Wang and Yun Liang",
year = "2018",
month = "1",
day = "1",
language = "English (US)",
note = "6th International Conference on Learning Representations, ICLR 2018 ; Conference date: 30-04-2018 Through 03-05-2018",

}

TY - CONF

T1 - Efficient recurrent neural networks using structured matrices in FPGAS

AU - Li, Zhe

AU - Wang, Shuo

AU - Ding, Caiwen

AU - Qiu, Qinru

AU - Wang, Yanzhi

AU - Liang, Yun

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Recurrent Neural Networks (RNNs) are becoming increasingly important for time series-related applications which require efficient and real-time implementations. The recent pruning based work ESE (Han et al., 2017) suffers from degradation of performance/energy efficiency due to the irregular network structure after pruning. We propose block-circulant matrices for weight matrix representation in RNNs, thereby achieving simultaneous model compression and acceleration. We aim to implement RNNs in FPGA with highest performance and energy efficiency, with certain accuracy requirement (negligible accuracy degradation). Experimental results on actual FPGA deployments shows that the proposed framework achieves a maximum energy efficiency improvement of 35.7× compared with ESE.

AB - Recurrent Neural Networks (RNNs) are becoming increasingly important for time series-related applications which require efficient and real-time implementations. The recent pruning based work ESE (Han et al., 2017) suffers from degradation of performance/energy efficiency due to the irregular network structure after pruning. We propose block-circulant matrices for weight matrix representation in RNNs, thereby achieving simultaneous model compression and acceleration. We aim to implement RNNs in FPGA with highest performance and energy efficiency, with certain accuracy requirement (negligible accuracy degradation). Experimental results on actual FPGA deployments shows that the proposed framework achieves a maximum energy efficiency improvement of 35.7× compared with ESE.

UR - http://www.scopus.com/inward/record.url?scp=85070971140&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85070971140&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:85070971140

ER -