Title :
Mapping Trained Neural Networks to FPNNs
Author :
Krcma, Martin ; Kastil, Jan ; Kotasek, Zdenek
Author_Institution :
Fac. of Inf. Technol., Brno Univ. of Technol., Brno, Czech Republic
Abstract :
This paper introduces a set of methods for mapping the trained neural networks into the lighted grid structured Field Programmable Neural Networks without the use of a training data set. These methods use information obtained from original neural networks such as a network structure, connection weights and biases. The principles of these mapping methods are described and the used grid FPNNs are explained. The results of experiments are presented and summarized.
Keywords :
field programmable gate arrays; neural nets; FPNN; connection weights; lighted grid structured field programmable neural networks; mapping methods; network structure; trained neural networks; Accuracy; Approximation methods; Biological neural networks; Field programmable gate arrays; IP networks; Neurons; FPGA; FPNA; FPNN; approximation; neural networks;
Conference_Titel :
Design and Diagnostics of Electronic Circuits & Systems (DDECS), 2015 IEEE 18th International Symposium on
Conference_Location :
Belgrade
Print_ISBN :
978-1-4799-6779-7
DOI :
10.1109/DDECS.2015.50
