Título: Uso de um Ambiente Codesign para a Implementação de Redes Neurais
Autores: Molz, Rolf F.; Engel, Paulo M.; Moraes, Fernando G.
Resumo: This paper presents a hardware implementation of neural networks into reprogrammable devices, FPGAs. It was used for this implementation the LIRMM board, which allows to develop both hardware and software designs, using C and VHDL languages, respectively. A feedforward neural network was implemented, using backpropagation learning. Emphasis is given for the net calculus, since intensive multiplications and additions are needed to compute this function. At the neuron level, the size of arithmetic operators was reduced by correctly coding the synaptics weights and input patterns, as well as replacing multipliers by adders and look-up-tables. At the network level, two techniques were employed to neuron synchronization. The synchronous solution resulted in a correct neural network implemented in hardware. After this implementation, the simulation of the neural network is presented. This work is organized as follows: section 1 introduces the use of neural network; section 2 presents the hardware-software environment used; section 3 details the neural network implementations (digital, analog, hybrid), our network and the simulation; and finally, our conclusions and future work.
Artigo em pdf: 4cbrn_006.pdf
Arquivo BibTex: 4cbrn_006.bib