Neural network clipping method, device and apparatus, and storage medium

The embodiment of the invention provides a method, equipment and device for clipping a neural network, and a storage medium. The method comprises the steps of obtaining an output value of each network layer module in a first neural network model based on the trained first neural network model and a corresponding data set; calculating a corresponding Laplacian matrix of each graph and a manifold measurement value of each network layer module based on the output value of each network layer module; calculating each manifold measurement distance between each non-last network layer module and the last network layer module based on the manifold measurement value of each network layer module; and obtaining a smaller manifold metric distance in the manifold metric distances, and connecting a non-last network layer module corresponding to the smaller manifold metric distance with a module behind the last network layer module to form a clipped second neural network model. In this way, the calculation amount and calcula