NEURAL NETWORK MODEL COMPRESSION METHOD AND APPARATUS, AND COMPUTER DEVICE

摘要

The present application provides a neural network model compression method and apparatus, and a computer device. The neural network model compression method provided in the present application comprises: decomposing, for each of original convolution layers of a neural network model to be compressed, the original convolution layers into a plurality of cascaded target convolution layers; acquiring a first convolution processing result after the original convolution layers perform convolution processing of inputted data, and a second convolution processing result after the plurality of cascaded target convolution layers perform convolution processing of the inputted data sequentially; correcting, according to the first convolution processing result and the second convolution processing result, weight matrixes of the plurality of cascaded target convolution layers, so as to obtain weight matrixes after the original convolution layers are compressed; and obtaining a compressed neural network model according to the weight matrixes after the original convolution layers are compressed.