Neural network search method and apparatus

The present disclosure relates to a neural network search method and apparatus. The method comprises: obtaining a neural network database to be searched and a training data set; ranking the neural networks, of which the number of training cycles is equal to a first preset value, in said neural network database in the descending order of the recognition accuracy of the training data set to obtain a first neural network sequence set, and forming, using the first M neural networks in the first neural network sequence set, a set of first neural networks to be trained; performing a first-stage training on said set of first neural networks using the training data set, the number of training cycles of the first-stage training being equal to a second preset value; and using the neural network, of which the number of training cycles is equal to the sum of the first preset value and the second preset value, in said neural network database as a target neural network. Also disclosed is a corresponding apparatus. The tech