Efficient presentations of learning samples to accelerate the convergence of learning in multilayer perceptron

Four methods of presenting learning samples are proposed to increase efficiency of learning in multilayer perceptrons. The methods involve presenting samples selectively instead of randomly; typical and confusing samples are selected and presented in systematic order. The methods were simulated to examine their effectiveness in a simple three-layer perceptron with two inputs and two outputs. All the methods except the method of presenting typical samples alone turned out to be superior to the conventional method of random presentation. The two best methods were to present typical samples in the first half period of learning and confusing ones in the second half period of learning, and to present in turn both typical and confusing samples