A Global Average Pooling Circuit for Convolutional Neural Network Computation

Abstract

The invention discloses a global average pooling circuit for convolutional neural network calculation, including an accumulator and a shift register, the accumulator is used to accumulate the convolution operation results of each layer in the convolution layer, the The shift register is used to perform a shift operation on the accumulation result of the accumulator, and output a reset signal to the accumulator after the shift is completed. When calculating the global average pooling layer operation, the data output in the accumulator is replaced by a simple shift operation to obtain the calculation result of the global average pooling layer, which can effectively improve the utilization of computing resources and speed up the overall system. calculation speed.

Description

technical field [0001] The invention belongs to the field of artificial intelligence algorithm hardware implementation, and more specifically relates to a global average pooling circuit used for convolutional neural network calculation. Background technique [0002] With the continuous development of artificial intelligence algorithms, various network models emerge in an endless stream, among which the convolutional neural network (CNN) is widely used for its unique weight sharing characteristics and higher prediction accuracy. [0003] The CNN model includes convolutional layers, pooling layers, and fully connected layers. Among them, the convolutional layer is the core of the CNN model, which extracts features from the input information of the input layer, the pooling layer performs feature selection and information filtering, and the fully connected layer further nonlinearly combines the extracted feature information to obtain output. In the traditional CNN model, the nu...

AI Technical Summary

Problems solved by technology

The traditional global average pooling circuit has no special processing method. When calculating the average value of the convolutional layer output, it is first necessary to write the convolution calculation result into the memory, then take it out of the memory, and then perform accumulation and division operations, resulting in The hardware resource overhead of CNN increases, and the calculation speed and throughput decrease

Images