Handwritten Digit Recognition System on an FPGA
Document Type
Conference Proceeding
Publication Date
1-8-2018
Publication Title
2018 IEEE 8th Annual Computing and Communication Workshop and Conference
Publisher
IEEE
Publisher Location
Las Vegas, NV
Volume
2018
First page number:
402
Last page number:
407
Abstract
This paper describes our implementation of a multilayer perceptron (MLP) learning network on a Cyclone IVE field programmable gate array (FPGA). The MLP uses MNIST data, the Modified National Institute of Standards and Technology database of handwritten digits, to train and test the design. Working with 8-bit precision, the FPGA design has similar accuracy and execution time as the 32-bit software solution but with 144 times slower clock frequency. With power consumption being proportional to frequency, the hardware solution provides power savings at no cost in accuracy or performance. Further reducing the precision from 8 to 4 bits only reduces accuracy from 89% to 78%, with area decreasing by 41%. Thus, the FPGA implementation of the MLP learning network offers a high-performance, low power alternative to traditional software methods.
Keywords
MNIST; FPGA; MLP; Deep learning; Machine learning; Hardware acceleration
Disciplines
Computer Sciences
Language
English
Repository Citation
Si, J.,
Harris, S. L.
(2018).
Handwritten Digit Recognition System on an FPGA.
2018 IEEE 8th Annual Computing and Communication Workshop and Conference, 2018
402-407.
Las Vegas, NV: IEEE.
http://dx.doi.org/10.1109/CCWC.2018.8301757