Article ID: | iaor1999883 |
Country: | Japan |
Volume: | J81-A |
Issue: | 3 |
Start Page Number: | 377 |
End Page Number: | 388 |
Publication Date: | Mar 1998 |
Journal: | Transactions of Institute of Electronics, Information, Communications |
Authors: | Tokuda Isao, Tamura Aki, Tokunaga Ryuji, Aihara Kazuyuki, Nagashima Tomomasa |
Keywords: | search, programming: nonlinear, neural networks |
A learning algorithm is introduced for chaotic dynamical systems which solve nonlinear optimization problems. The algorithm controls the asymptotic measure of the chaotic dynamical system and improves an efficiency of the ‘chaotic search’ dynamics for optimum solution. Using several instances of 1- and 2-dimensional nonlinear optimization problems, performance of the learning algorithm is demonstrated. It is also shown that the learning algorithm works as a ‘chaotic simulated annealing’, which realizes a gradual convergence of the ‘chaotic search’ dynamics to possible optimum solution.