Adaptive and reconfigurable deep convolutional neural network computing method and device

Abstract

The invention relates to an adaptive and reconfigurable deep convolutional neural network computing method and device. The method comprises the steps that a program execution process of the computing device is decided according to a control signal; dynamic reconfiguration is performed on basic computing elements according to deep convolutional neural network size parameters to determine a combination level and a parallelism degree of arithmetic units; and corresponding processed data is loaded according to different reconfiguration conditions, corresponding computing is performed on convolutional neural network layers of different properties, and finally a connection output result of a group of neurons is obtained. Through the computing method and device, the defect of low flexibility of special hardware is overcome, and design parameters of the arithmetic units can be reconfigured to achieve the purpose of supporting deep convolutional neural networks of different sizes; parallel operation of convolution kernels of the same size can be realized, parallel operation of convolution kernels of different sizes can also be realized, the parallelism degree of deep convolutional neural network operation is greatly improved through the arithmetic units which can be dynamically reconfigured, and computing performance is improved.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to an adaptive and reconfigurable deep convolutional neural network calculation method and device. Background technique [0002] This section introduces readers to background technologies that may be related to various aspects of the present invention, and it is believed that useful background information can be provided to readers, thereby helping readers to better understand various aspects of the present invention. Accordingly, it is to be understood that the descriptions in this section are for the purposes stated above and do not constitute admissions of prior art. [0003] Deep neural networks have produced excellent results in many current application fields, such as face recognition, object detection, automatic driving, speech recognition, etc., and have been widely used. With the improvement of the accuracy of the algorithm, the depth of the neural network is increasing,...

AI Technical Summary

Problems solved by technology

The computing primitives with a fixed design can only calculate these convolution operations of different scales in series, and the degree of parallelism is low, which is not enough to meet the computing performance requirements of practical applications.

Images