Summary

Proceedings of the 2012 International Symposium on Nonlinear Theory and its Applications

2012

Session Number:C3L-B

Session:

Number:723

A Modular Neural Network for Parallel Computation

Yoshihiro Hayakawa,  Daisuke Sasaki,  Koji Nakajima,  

pp.723-726

Publication Date:

Online ISSN:2188-5079

DOI:10.15248/proc.1.723

PDF download (945.7KB)

Some of combinatorial optimization problems cause exponential increases of calculation time in terms of a problem size. In case of using a neural network solver, faster calculation is prospective because each neuron is updated in essentially parallel. In fact, however, a neural network solver is performed by using the ordinary differential equation in a computer, so that a calculation speed is not enough yet. Hence, in this paper we propose a modular neural network for implementing parallel computation with a MPI technique. Each module is assigned to each processor and these are calculated in parallel in our proposed modular neural network and we discuss the possibility of improvement of calculation by decreasing communication frequency between processors.

[1] Y. Hayakawa and K. Nakajima, “Design of the inverse function delayed neural network for solving combinatorial optimization problems”, IEEE Trans. Neural Netw., vol. 21, no. 2, pp. 224-237, 2010.

[2] T. Sota, Y. Hayakawa, S. Sato and K. Nakajima, “An application of higher order connection to inverse function delayed network”, Nonlinear Theory and Its Applications, IEICE, vol. 2, no. 2, pp. 180-197, 2011.

[3] Peter S. Pacheco, “PALALLEL PROGRAMMING with MPI”, Morga Kaufmann Publishers, 1997.