Neural network training method and device

Abstract

The invention provides a neural network training method and a neural network training device. The neural network training device comprises a selection unit and an operation unit, wherein the operationunit is used for training a neural network. The neural network training method and the neural network training device provided by the invention have the advantages of low cost and low energy consumption.

Description

technical field [0001] The present application relates to the technical field of chip processing, in particular to a neural network training method and device. Background technique [0002] Neural networks have been used very successfully. But as we design larger and deeper neural networks, more weights will be introduced, and super-large weights become a huge challenge for neural network applications. Large-scale weights also put forward higher requirements on the computing unit, and the computing time and computing energy consumption increase accordingly. Therefore, reducing the weight of the neural network and reducing the calculation time has become an urgent problem to be solved. [0003] At present, most of the work mainly uses the method of pruning to cut the neurons or weights of the neural network, thereby reducing the scale of the neural network parameters, thereby reducing the calculation time. However, the sparse neural network after pruning will introduce irr...

AI Technical Summary

Problems solved by technology

However, the sparse neural network after pruning will introduce irregularities, which will make the topology of the original rules irregular, making it difficult for the processing platform to obtain performance improvements.

Images