A Convolutional Neural Network Pooling Unit Design Method

Abstract

The invention discloses a method for designing a convolutional neural network pooling unit, which relates to the field of neural network technology implementation; a design system for a convolutional neural network pooling unit is established, wherein the feature map and parameter input modules are used to cache feature maps and parameters to be pooled. The pooling parameters that need to be configured, the parameter parsing and mapping module receives the pooling parameters from the feature map and the parameter input module, parses the pooling parameters, configures the register group according to the parsed parameters, and constructs the state group in the state group module According to the jump state of the state machine, the state unit module adopts the method of circuit multiplexing and pooling parameter fusion, calls the pooling calculation module to realize the pooling calculation of different pooling parameters, and outputs the result to the pooling A result output module; using the system to design a convolutional neural network pooling unit.

Description

technical field [0001] The invention discloses a unit design method, relates to the field of neural network technology realization, in particular to a convolutional neural network pooling unit design method. Background technique [0002] With the development of the artificial intelligence (AI) field, the convolutional neural network (CNN) has been fully utilized. The current mainstream convolutional neural network model not only has a complex structure, but also has a large amount of calculation data, and the architecture of each layer is also very different. It is not easy to achieve high performance while achieving high versatility. Both resource utilization and energy efficiency ratio must be considered. The pooling layer is generally connected after the convolutional layer, which is the secondary feature extraction of the feature map, which can reduce the resolution of the feature map, reduce the data scale, and simplify the network structure. The pooling operation is a...

AI Technical Summary

Problems solved by technology

There are big differences in the pooling calculations of different network structures and different layers of the same network structure

Images