Sparse convolutional neural network sorting method, operation method and device and equipment

Abstract

The invention discloses a sparse convolutional neural network sparse data sorting method, an operation method and device, a storage medium and equipment. According to the scheme, the method comprisesthe steps: obtaining to-be-processed feature map data and convolution kernel data after sorting processing; obtaining a mark sequence of a first weight vector of the sorted convolution kernel data; obtaining a first feature vector in the to-be-processed feature map data, wherein the first feature vector and the first weight vector are subjected to multiply-add operation; and sorting the feature values of the first feature vector according to the mark sequence, and deleting a feature value matched with the zero weight value removed in the zero weight value removing process to obtain a second feature vector. The convolution kernel data and the feature map data are compressed in the channel direction based on the first weight vector and the second feature vector, so that the data volume of convolution operation is greatly reduced, the operation speed of hardware for a sparse convolutional neural network is increased, and the waste of hardware performance and computing resources is avoided.

Description

technical field [0001] The invention relates to the technical field of data processing, in particular to a sorting method, operation method, device and equipment of a sparse convolutional neural network. Background technique [0002] Deep learning (Deep learning) is one of the important application technologies of AI (Artificial intelligence, artificial intelligence), which is widely used in computer vision, speech recognition and other fields. Among them, CNN (Convolutional Neural Network, Convolutional Neural Network) is a deep learning and efficient recognition technology that has attracted attention in recent years. Convolution operations and vector operations to produce high accuracy results in image and speech recognition. [0003] However, with the development and wide application of the convolutional neural network, it faces more and more challenges. For example, the parameter scale of the CNN model is getting larger and larger, which makes the CNN model have a very...

AI Technical Summary

Problems solved by technology

If these sparse data structures are directly calculated on the hardware, it will waste hardware performance and computing resources, making it difficult to improve the computing speed of the CNN model.

Images