Abstract: First analyzed is the pattern space from each neuron of binary feedforward perceptron. Then a rapid, reliable and available learning algorithm using the characteristics of the inner product and area of attraction. In relation to threshold of neuron with pattern direction of the inner product, the neural network is trained to have capacity of controllable error correction,with simple structure of the neural network and easiness to be implemented by hardware. Finally this strategy is demonstrated to have good performance when used for pattern recognition and Boolean function approximation.

Key words: neural networks, learning algorithm, inner product, pattern recognition, Booleanfunction