Implementation of single neuron using various activation functions with FPGA

Author

Jeyanthi, S. ; Subadra, M.

Author_Institution

Dept. of Electron. & Commun. Eng., Einstein Coll. of Eng., Tirunelveli, India

fYear

2014

fDate

8-10 May 2014

Firstpage

1126

Lastpage

1131

Abstract

A main characteristic element of any artificial neural network (ANN) is an activation function since their results are used as a starting point for any complex ANN application. This paper is about design and implementation of single neuron with three different types of activation function sigmoid, linear and threshold activation function. Activation function package is written using VHDL and single precision floating point representation method is used to design the single neuron. The simulation result is taken using Xilinx ISE14.5i and the performance of different activation functions are analyzed and compared in terms of device utilization and CPU time.

Keywords

field programmable gate arrays; floating point arithmetic; hardware description languages; neural nets; transfer functions; ANN; CPU time; FPGA; VHDL; Xilinx ISE14.5i; activation function package; activation function sigmoid; artificial neural network; device utilization; linear activation function; single neuron; single precision floating point representation method; threshold activation function; Approximation methods; Computers; Conferences; Field programmable gate arrays; Hardware; Mathematical model; Neurons; Artificial neuron; FPGA VHDL; activation function;

fLanguage

English

Publisher

ieee

Conference_Titel

Advanced Communication Control and Computing Technologies (ICACCCT), 2014 International Conference on

Conference_Location

Ramanathapuram

Print_ISBN

978-1-4799-3913-8

Type

conf

DOI

10.1109/ICACCCT.2014.7019273

Filename

7019273