Scribd
Upload
180 views

Implementation of Neural Network Back Propagation Training Algorithm On FPGA

This document discusses the implementation of neural network backpropagation training algorithms on field-programmable gate arrays (FPGAs). It begins with an introduction to artificial neural networks and the backpropagation algorithm. It then describes using a 2:2:2:1 neural network on an FPGA to implement basic digital gates, finding the hyperbolic tangent sigmoid transfer function gave the best error convergence. A 4:4:2:4 network was also used for image compression. In conclusion, the backpropagation algorithm was successfully implemented on an FPGA, which could enable online neural network training on FPGAs. Further optimization and use of adaptive learning rates could improve hardware efficiency and training time.

Uploaded by

Mix Ruksin
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
180 views

Implementation of Neural Network Back Propagation Training Algorithm On FPGA

This document discusses the implementation of neural network backpropagation training algorithms on field-programmable gate arrays (FPGAs). It begins with an introduction to artificial neural networks and the backpropagation algorithm. It then describes using a 2:2:2:1 neural network on an FPGA to implement basic digital gates, finding the hyperbolic tangent sigmoid transfer function gave the best error convergence. A 4:4:2:4 network was also used for image compression. In conclusion, the backpropagation algorithm was successfully implemented on an FPGA, which could enable online neural network training on FPGAs. Further optimization and use of adaptive learning rates could improve hardware efficiency and training time.

Uploaded by

Mix Ruksin
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 19

Implementation of Neural Network Back

Propagation Training Algorithm on FPGA


Presented by

Ruksin Sangrugee

Abstract
using software tools in the computer system

Off-line Method
fixed and lack of getting trained during usage

On-line Method

trainable Artificial Neural Network (ANN) chip

Introduction

As you read these word, you are using a


complex biological neural network.
1011 neurons
to facilitate our reading, breathing, motion and thinking

In order to implement a function using


ANN, we use back propagation algorithm
in its gradient form because it is
computation and hardware efficient

The limitation in the implementation of neural network


on FPGA* is the number of multipliers

*A

field-programmable gate array (FPGA) is an integrated circuit designed to be


configured by a customer or a designer after manufacturing.

Back Propagation Training Algorithm

is a supervised learning algorithm for


multilayer feed forward neural network.

Training Neural Network Using Back Propagation


Algorithm to Implement Basic Digital Gates

The Figure 2 shows the architecture of


2:2:2:1 neural network selected to
implement basic digital gates.

MATLAB analysis is done on the neural


network architecture and it is found that
TANSIG (hyperbolic tangent sigmoid)
transfer function gives the best
convergence of error.

Simulation Results

The 2:2:2:1 network with training algorithm


consumes 81% of area on SPARTAN3
FPGA

Xilinx SPARTAN FPGA

Training Neural Network Using Back Propagation


Algorithm to Implement Image Compression

The input image is normalized with the


maximum pixel value, for grey scale image
with grey levels [0,255] in the range [0,1]

Reson:
is due to the fact that neural networks can operate more efficiently
when both their inputs and outputs are limited to a range of [0,1].

This also avoids use of larger numbers in


the working if the network, which reduces
the hardware requirement drastically when
implemented on FPGA

For the selected 4:4:2:4 network,


MATLAB analysis is done using various
transfer function like PURELIN, LOGSIG,
and TANSIG.

The network is trained using MATLAB


neural network toolbox back propagation
training function TRAINGD

The general structure for 4:4:2:2:4 artificial


neural network with online training is
shown in Figure 7.

Conclusion and Future Score

The back propagation algorithm for training


multilayer artificial neural network is studied
and successfully implemented on FPGA.

This can help achieving online training of


neural network on FPGA.

In term of hardware efficiency, the design


can be optimized to reduce the number of
multipliers.

The training time for the network to


converge to the expected output can be
reduced by implement the adaptive
learning rate back-propagation algorithm.

References
International Journal of Computer Application (0975 - 8887)
Volume 52 - No.6, August 2012

You might also like