FPGA-based Deep-Learning Accelerators for Energy Efficient Motor Imagery EEG classification

被引:1
作者
Flood, Daniel [1 ]
Robinson, Neethu [2 ]
Shreejith, Shanker [1 ]
机构
[1] Trinity Coll Dublin, Dept Elect & Elect Engn, Dublin, Ireland
[2] Nanyang Technol Univ, Sch Comp Sci & Engn, Singapore, Singapore
来源
2022 IEEE INTERNATIONAL CONFERENCE ON OMNI-LAYER INTELLIGENT SYSTEMS (IEEE COINS 2022) | 2022年
关键词
Brain-Computer Interfaces; Deep Learning; Field Programmable Gate Arrays; AI Accelerators; BRAIN-COMPUTER INTERFACES; COMMUNICATION;
D O I
10.1109/COINS54846.2022.9854985
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
In recent years, Deep Learning has emerged as a powerful framework for analysing and decoding bio-signals like Electroencephalography (EEG) with applications in brain computer interfaces (BCI) and motor control. Deep convolutional neural networks have shown to be highly effective in decoding BCI signals for applications like two-class motor imagery decoding. Their deployment in real-time applications, however, requires highly parallel and capable computing platforms like GPUs to achieve high-speed inference, consuming a large amount of energy. In this paper, we explore a custom deep learning accelerator on an off-the-shelf hybrid FPGA device to achieve similar inference performance at a fraction of the energy consumption. We evaluate different optimisations at bit-level, data-path and training using a state-of-the-art deep convolutional neural network as our baseline model to arrive at our custom precision quantised deep learning model, which is implemented using the FINN compiler from Xilinx. The accelerator, deployed on a Xilinx Zynq Ultrascale+ FPGA, achieves a significant reduction in power consumption (approximate to 17x), sub 2 ms decoding latency and a near-identical decoding accuracy (statistically insignificant reduction of 2.5% average) as the reported baseline subject-specific classification accuracy on an N (= 54) subject motor imagery EEG (MI-EEG) dataset compared to the Deep CNN model on GPU, making our approach more appealing for low-power real-time BCI applications. Furthermore, this design approach is transferable to other deep learning models reported in BCI research, paving the way for novel applications of real-time portable BCI systems.
引用
收藏
页码:325 / 330
页数:6
相关论文
共 32 条
  • [1] Bashivan P, 2016, Arxiv, DOI [arXiv:1511.06448, DOI 10.48550/ARXIV.1511.06448]
  • [2] A Hardware/Software Prototype of EEG-based BCI System for Home Device Control
    Belwafi, Kais
    Ghaffari, Fakhreddine
    Djemal, Ridha
    Romain, Olivier
    [J]. JOURNAL OF SIGNAL PROCESSING SYSTEMS FOR SIGNAL IMAGE AND VIDEO TECHNOLOGY, 2017, 89 (02): : 263 - 279
  • [3] Cecotti H., 2009, ESANN
  • [4] Echtioui A., 2021, 2021 INT WIR COMM MO, P1648, DOI [10.1109/IWCMC51323.2021.9498833, DOI 10.1109/IWCMC51323.2021.9498833]
  • [5] Inter-subject transfer learning with an end-to-end deep convolutional neural network for EEG-based BCI
    Fahimi, Fatemeh
    Zhang, Zhuo
    Goh, Wooi Boon
    Lee, Tih-Shi
    Ang, Kai Keng
    Guan, Cuntai
    [J]. JOURNAL OF NEURAL ENGINEERING, 2019, 16 (02)
  • [6] Han S, 2016, Arxiv, DOI arXiv:1510.00149
  • [7] He B, 2015, P IEEE, V103, P907, DOI [10.1109/JPROC.2015.2407272, 10.1109/jproc.2015.2407272]
  • [8] Heelan C, 2018, IEEE ENG MED BIO, P1070, DOI 10.1109/EMBC.2018.8512415
  • [9] Deep learning
    LeCun, Yann
    Bengio, Yoshua
    Hinton, Geoffrey
    [J]. NATURE, 2015, 521 (7553) : 436 - 444
  • [10] Lee KS, 2017, Arxiv, DOI [arXiv:1706.00120, 10.48550/arXiv.1706.00120, DOI 10.48550/ARXIV.1706.00120]