Sparse neural network calculation method and device based on systolic array

A pulsating array and neural network technology, applied in the field of artificial intelligence, can solve the problems of large resource consumption and achieve the effects of sufficient data reuse, reduced loss, and high efficiency

Active Publication Date: 2020-11-13
南京风兴科技有限公司
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] This application provides a sparse neural network calculation method based on a systolic array to solve the problems of low efficiency in the encryption and decryption process and large resource consumption in the prior art

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Sparse neural network calculation method and device based on systolic array
  • Sparse neural network calculation method and device based on systolic array
  • Sparse neural network calculation method and device based on systolic array

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0049] see figure 1 , is a flow chart of a sparse neural network computing method based on a systolic array architecture in this application.

[0050] Depend on figure 1 It can be seen that a sparse neural network calculation method based on a systolic array architecture provided by the embodiment of the present application includes:

[0051] S100: Obtain a feature map containing n weights; the size of the feature map is x*y;

[0052] In this embodiment, the acquired feature map is derived from the data input into the computing architecture at the previous stage, and n weights represent the number of sub-feature blocks to be divided into the feature map, so that after the sign map is divided into blocks, each sub-feature block corresponds to a weight. The feature map can usually be expressed in the form of x*y, such as 64*8, 128*8 and so on.

[0053] S200: Divide the feature map into n sub-feature blocks on average along the x-axis direction, and the size of each sub-featu...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a sparse neural network calculation method based on a systolic array. The method comprises the steps of obtaining a feature map containing n weights, wherein the size of the feature map is x * y; averagely segmenting the feature map into n sub-feature blocks along the x-axis direction, wherein the size of each sub-feature block is (x/n) * y; wherein each sub-feature block corresponds to one weight; calculating each sub-feature block according to the position of the weight in the weight matrix to obtain a calculation result; and regenerating a weight matrix according tothe calculation result and outputting the weight matrix. According to the invention, sparse convolution calculation is realized in a pulsation array manner, so that data multiplexing is more sufficient, convolution is carried out in a blocking manner, calculation is more flexible and efficient, the weight is input to lower-level equipment after being coded, only a non-zero weight is input to the architecture for calculation, and the loss of a coding unit is reduced.

Description

technical field [0001] The present application relates to the technical field of artificial intelligence, in particular to a method and device for calculating a sparse neural network based on a systolic array. Background technique [0002] With the continuous development of artificial intelligence (AI), it has evolved from early artificial feature engineering to learning from massive data, and can be applied to many fields such as machine vision, speech recognition, and natural language processing. Convolutional Neural Network (CNN) is one of the most representative network structures in deep learning technology. It is becoming more and more popular in the field of artificial intelligence, especially in the field of image processing, and has achieved remarkable results. [0003] As the network coverage becomes wider and more complex, the computing resources used for convolution training also grow exponentially, and for neural networks with more layers and nodes, it becomes c...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06N3/04
CPCG06N3/045
Inventor 陶为王中风刘文剑谢逍茹
Owner 南京风兴科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap