Device and method for performing pooling operation

Abstract

The invention discloses a device for performing pooling operation. The device comprises an instruction storage unit, a controller unit, a data access unit and an operation module. The instruction storage unit reads in instructions through the data access unit and buffers the read instructions; the controller unit reads the instructions from the instruction storage unit and decodes the instructions into control signals for controlling the behavior of an operation module and then distributes the control signals to the operation module; the data access unit is used for accessing external address space and completing data loading and storing; the operation module is used for completing maximum value seeking operation of the maxpooling operation or accumulating and multiplying operation of the avgpooling operation. For the maxpooling, in the forward operation, the operation module circularly reads input vectors of the poling kernel and performs comparison of size to obtain new output vectors of the kernel and save an index vector corresponding to each output vector at the same time until the pooling operation of the layer is completed. The invention can solve the problems of insufficient computing performance of CPU and GPU and high front-end decoding cost.

Description

technical field [0001] The present invention relates to an artificial neural network, in particular to a device and method for performing pooling operations. Background technique [0002] Multi-layer artificial neural networks are widely used in the fields of pattern recognition, image processing, function approximation, and optimization calculations. In recent years, multi-layer artificial networks have been favored by academic circles and researchers due to their high recognition accuracy and good parallelism. industry is getting more and more attention. [0003] The pooling operation refers to the downsampling operation of local features in the feature layer of the neural network to reduce the dimension of the feature layer. There are two types of pooling operations: maxpooling refers to taking the maximum value as the result in the kernel; avgpooling refers to taking the average value as the result in the kernel. The kernel here is the pooling core, the size is specifi...

AI Technical Summary

Problems solved by technology

Since the GPU is a device specially used to perform graphics and image calculations and scientific calculations, without special support for multi-layer artificial neural network operations, it still requires a lot of front-end decoding work to perform multi-layer artificial neural network operations, which brings a lot of problems. additional cost

In addition, the GPU only has a small on-chip cache, and the model data (weights) of the multi-layer artificial neural network need to be repeatedly moved from off-chip, and the off-chip bandwidth has become the main performance bottleneck.

In addition, the GPU has only a small on-chip cache, and the model data (weights) of the multi-layer artificial neural network need to be repeatedly moved from off-chip. The off-chip bandwidth has become the main performance bottleneck, and it has brought huge power consumption overhead.

Images