Abstract:
By introducing quantum superposition into neural networks, this paper proposes a model of quantum neural network, in which the units in the hidden and output layers adopt the superposition of multilevel activation functions to obtain the multi-lever partitions of the feature space. During the training, the units can be `collapsedin' and `spreadout' according to practical requirement. Moreover, this proposed algorithm can learn the inaccuracy and uncertainties from the input of fuzzy information. It is shown from the application for ECG classification that the proposed algorithm has faster training speed and better performance.