Scalable Architecture for on-Chip Neural Network Training using Swarm Intelligence

Abstract

This paper presents a novel architecture for on-chip neural network training using particle swarm optimization (PSO). PSO is an evolutionary optimization algorithm with a growing field of applications which has been recently used to train neural networks. The architecture exploits PSO algorithm to evolve network weights as well as a method called layer partitioning to implement neural networks. In the proposed method, a neural network is partitioned into groups of neurons and the groups are sequentially mapped to available functional units. Thus, the architecture is reconfigurable for training and implementing different multilayer feedforward neural networks without the need for modifying the architecture. The implementation is intended for real-time applications regarding hardware cost and speed. The results show that the proposed system provides a trade-off between resource requirements and speed.

Fpga Implementation of Ann Training Using Levenberg and Marquardt Algorithms

“…The studies related to implementation of an ANN training and/or a trained ANN on FPGA are presented in the literature. Ferrer, Gonzalez, Fleitas, Acle and Canetti [9], Savich, Moussa and Areibi [10], Farmahini-Farahani, Fakhraie and Safari [11] have used fixed-point number format at various bit lengths. Nedjah, Silva, Mourelle and Silva [12], Ç avuşlu, Karakuzu and Şahin [13], Ç avuşlu, Karakuzu, Şahin and Yakut [14], Ç avuşlu, Karakuzu and Karakaya [15] have used floatingpoint number format at various bit-lengths.…”

“…Nedjah, Silva, Mourelle and Silva [12], Won [16], Ç avuşlu, Karakuzu and Şahin [13], Ç avuşlu, Karakuzu, Şahin and Yakut [14], Ç avuşlu, Karakuzu and Karakaya [15], Savich, Moussa and Areibi [10] have used logarithmic sigmoidal approaches as activation function. Farmahini-Farahani, Fakhraie and Safari [11], Ferrer, Gonzalez, Fleitas, Acle and Canetti [9], Farmahini-Farahani, Fakhraie and Safari [11], Ç avuşlu, Karakuzu and Karakaya [15] have used tangent hyperbolic activation function approaches as activation function.…”

“…Won [16],Farmahini-Farahani, Fakhraie and Safari [11] have used look-up table approach for hardware implementations of activation functions. Ferreira, Ribeiro, Antunes and Dias [17], Ferrer, Gonzalez, Fleitas, Acle and Canetti [9], Savich, Moussa and Areibi [10] have used piecewise linear approach.Nedjah, Silva, Mourelle and Silva [12] have used parabolic approach.…”

“…Savich, Moussa and Areibi [10], Ç avuşlu, Karakuzu, Şahin and Yakut [14] have used back-propagation algorithm for network training. Farmahini-Farahani, Fakhraie and Safari [11], Ç avuşlu, Karakuzu and Karakaya [15] have used PSO algorithm in network training.…”

FPGA Implementation of Wavelet Neural Network Training with PSO/iPSO

“…al. 22 implemented ANN training based on FPGA using the PSO algorithm. In the application fixed point number ws-jcsc format is used as the number format and the look-up table is used for the tangent hyperbolic activation function.…”