IMPLEMENTATION OF A
FAST ARTIFICIAL NEURAL NETWORK LIBRARY (FANN)

 

STEFFEN NISSEN
lukesky@diku.dk

 
October 31, 2003

 
Department of Computer Science
University of Copenhagen (DIKU)

 
 

 

Abstract:

This report describes the implementation of a fast artificial neural network library in ANSI C called fann. The library implements multilayer feedforward networks with support for both fully connected and sparse connected networks. Fann offers support for execution in fixed point arithmetic to allow for fast execution on systems with no floating point processor. To overcome the problems of integer overflow, the library calculates a position of the decimal point after training and guarantees that integer overflow can not occur with this decimal point.

The library is designed to be fast, versatile and easy to use. Several benchmarks have been executed to test the performance of the library. The results show that the fann library is significantly faster than other libraries on systems without a floating point processor, while the performance was comparable to other highly optimized libraries on systems with a floating point processor.

Keywords: ANN, artificial neural network, performance engineering, fixed point arithmetic, ANSI C.