Area-Efficient FPGA Implementation of Minimalistic Convolutional Neural Network Using Residue Number System

2018 ◽  
Author(s):  
Nikolay I. Chervyakov ◽  
Pavel A. Lyakhov ◽  
Maria V. Valueva ◽  
Georgii V. Valuev ◽  
Dmitrii I. Kaplun ◽  
...  
Keyword(s):  
Neural Network ◽  
Convolutional Neural Network ◽  
Number System ◽  
Residue Number System ◽  
Fpga Implementation ◽  
Residue Number ◽  
Area Efficient

scholarly journals Hardware implementation of a convolutional neural network using calculations in the residue number system

2019 ◽  
Vol 43 (5) ◽  
pp. 857-868 ◽  
Author(s):  
N.I. Chervyakov ◽  
P.A. Lyakhov ◽  
N.N. Nagornov ◽  
M.V. Valueva ◽  
G.V. Valuev
Keyword(s):  
Neural Network ◽  
Convolutional Neural Network ◽  
Network Architecture ◽  
Hardware Implementation ◽  
Traditional Approach ◽  
Number System ◽  
Residue Number System ◽  
Quantization Noise ◽  
The Neural Network ◽  
Residue Number

Modern convolutional neural networks architectures are very resource intensive which limits the possibilities for their wide practical application. We propose a convolutional neural network architecture in which the neural network is divided into hardware and software parts to increase performance and reduce the cost of implementation resources. We also propose to use the residue number system in the hardware part to implement the convolutional layer of the neural network for resource costs reducing. A numerical method for quantizing the filters coefficients of a convolutional network layer is proposed to minimize the influence of quantization noise on the calculation result in the residue number system and determine the bit-width of the filters coefficients. This method is based on scaling the coefficients by a fixed number of bits and rounding up and down. The operations used make it possible to reduce resources in hardware implementation due to the simplifying of their execution. All calculations in the convolutional layer are performed on numbers in a fixed-point format. Software simulations using Matlab 2017b showed that convolutional neural network with a minimum number of layers can be quickly and successfully trained. Hardware implementation using the field-programmable gate array Kintex7 xc7k70tfbg484-2 showed that the use of residue number system in the convolutional layer of the neural network reduces the hardware costs by 32.6% compared with the traditional approach based on the two’s complement representation. The research results can be applied to create effective video surveillance systems, for recognizing handwriting, individuals, objects and terrain.

Residue Number System-Based Solution for Reducing the Hardware Cost of a Convolutional Neural Network

2020 ◽  
Vol 407 ◽  
pp. 439-453
Author(s):  
N.I. Chervyakov ◽  
P.A. Lyakhov ◽  
M.A. Deryabin ◽  
N.N. Nagornov ◽  
M.V. Valueva ◽  
...  
Keyword(s):  
Neural Network ◽  
Convolutional Neural Network ◽  
Number System ◽  
Residue Number System ◽  
Hardware Cost ◽  
Residue Number

scholarly journals CryptoDL: Predicting Dyslexia Biomarkers from Encrypted Neuroimaging Dataset Using Energy-Efficient Residue Number System and Deep Convolutional Neural Network

Symmetry ◽  
2020 ◽  
Vol 12 (5) ◽  
pp. 836 ◽  
Author(s):  
Opeyemi Lateef Usman ◽  
Ravie Chandren Muniyandi
Keyword(s):  
Neural Network ◽  
Convolutional Neural Network ◽  
Critical Path ◽  
Imaging Techniques ◽  
Brain Magnetic Resonance Imaging ◽  
Number System ◽  
Residue Number System ◽  
Deep Convolutional Neural Network ◽  
Path Delay ◽  
Residue Number

The increasing availability of medical images generated via different imaging techniques necessitates the need for their remote analysis and diagnosis, especially when such datasets involve brain morphological biomarkers, an important biological symmetry concept. This development has made the privacy and confidentiality of patients’ medical records extremely important. In this study, an approach for a secure dyslexia biomarkers classification is proposed using a deep learning model and the concept of residue number system (RNS). A special moduli set of RNS was used to develop a pixel-bitstream encoder that encrypts the 7-bit binary value of each pixel present in the training and testing brain magnetic resonance imaging (MRI) dataset (neuroimaging dataset) prior to classification using cascaded deep convolutional neural network (CNN). Theoretical analysis of our encoder design shows that the proposed pixel-bitstream encoder is a combinational circuit that requires fewer fast adders, with area complexity of 4n AFA and time delay of (3n + 3) DFA for n ≥ 3. FPGA implementation of the proposed encoder shows 23.5% critical path delay improvement and saves up to 42.4% power. Our proposed cascaded deep CNN also shows promising classification outcomes, with the highest performance accuracy of 73.2% on the encrypted data. Specifically, this study has attempted to explore the potencies of CNN to discriminate cases of dyslexia from control subjects using encrypted dyslexia biomarkers neuroimaging dataset. This kind of research becomes expedient owing to the educational and medical importance of dyslexia.

Sign in / Sign up

Export Citation Format

Share Document