Title :
Global optimization for fast multilayer perceptron training
Author_Institution :
Dept. of Ind. Eng., Pohang Sci. & Technol. Univ., South Korea
Abstract :
A new training algorithm for multilayer perceptrons (MLPs) is proposed. The proposed algorithm consists of two phases: a trust region-based local search for fast training of networks and a quotient-based global search for escaping local minima and moving toward a weight vector of next descent. These two phases are repeated alternatively in the weight space to achieve a goal training error. Benchmark results demonstrate a significant performance improvement of the proposed training algorithm compared with other existing training algorithms.
Keywords :
learning (artificial intelligence); multilayer perceptrons; optimisation; search problems; global optimization; goal training error; multilayer perceptron training; quotient-based global search; training algorithm; trust region-based local search; weight vector; Benchmark testing; Convergence; Gradient methods; Industrial engineering; Industrial training; Multilayer perceptrons; Optimization methods; Pattern recognition; Robotics and automation; Supervised learning;
Conference_Titel :
Neural Networks, 2003. Proceedings of the International Joint Conference on
Print_ISBN :
0-7803-7898-9
DOI :
10.1109/IJCNN.2003.1223381
