An efficient learning method for layered neural networks based on selection of training data and input characteristics of an output layer unit

This paper proposes an efficient learning method for a layered neural network based on the selection of training data and the input characteristics of an output layer unit. Compared to recent neural networks, pulse neural networks, and quantum neuro computation, the multilayer neural network is widely used due to its simple structure. When learning objects are complicated, problems such as unsuccessful learning or a significant time required in learning remain unsolved. The aims of this paper are to suggest solutions for these problems and to reduce the total learning time. The total learning time means the total computational time required to learn certain objects, including adjusting parameter values and restarting learning from the beginning. Focusing on the input data during the learning stage, we undertook an experiment to identify the data that create large errors and interfere with the learning process. Our method divides the learning process into several stages. In general, the input characteristics to an output layer unit show oscillation during the learning process for complicated problems. Focusing on the oscillatory characteristics, it is determined whether the learning will move on to the next stage or the learning will restart from the beginning. Computational experiments suggest that the proposed method has the capability for higher learning performance and needs less learning time compared with the conventional method. © 2012 Wiley Periodicals, Inc. Electron Comm Jpn, 95(4): 57–67, 2012; Published online in Wiley Online Library ( wileyonlinelibrary.com ). DOI 10.1002/ecj.10365