Convolution calculation device and method

Abstract

An embodiment of the invention provides a convolution calculation device which is applied to convolution calculation of a convolution layer and comprises a decompression circuit, a data combination circuit and a calculation circuit. The decompression circuit is used for decompressing compressed weight data of a convolution kernel of the convolution layer to generate decompressed weight data. The data combination circuit is used for combining the decompressed data with uncompressed weight data of the convolution kernel so as to restore the data sequence of the weight data of the convolution kernel. And the calculation circuit performs calculation according to the weight data of the convolution kernel and the input data of the convolution layer. Because the weight data of the convolution kernel which is compressed in advance is transmitted to the convolution calculation device, when convolution operation is carried out, the compressed weight data is firstly decompressed, and then convolution operation is carried out according to the decompressed weight data, so that the storage space occupied by the convolution kernel in the electronic equipment and the bandwidth required during transmission can be reduced.

Description

Technical field[0001]This application belongs to the field of artificial intelligence technology, and in particular relates to a data processing method, convolution calculation device and method.Background technique[0002]At present, more and more fields are beginning to apply artificial intelligence technology. In artificial intelligence technology, electronic devices can use training samples to perform machine learning on computing models (ie, neural network models). After that, the electronic device can use the trained calculation model to perform some tasks, such as classification and prediction. For example, Convolutional Neural Networks (CNN) can be used for image recognition. However, the data of the convolution kernel in some convolutional layers is huge, which requires more system resources when performing convolution processing of the convolutional layer on an electronic device.Summary of the invention[0003]The embodiments of the present application provide a convolution ca...

AI Technical Summary

Problems solved by technology

However, the data of the convolution kernel in some convolutional layers is huge, resulting in the need to occupy more system resources when performing convolution processing of the convolutional layer on electronic devices

Images