Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Convolutional neural network optimization method and device based on pulse array

A convolutional neural network and array technology, applied in the field of convolutional neural network optimization, can solve problems such as strict consistency, physical parameters of PE cannot be guaranteed, and the highest operating frequency cannot be strictly consistent.

Active Publication Date: 2020-02-28
JILIN UNIV
View PDF9 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] However, limited by the process technology, the physical parameters of many PEs cannot be strictly consistent, but are normally distributed, which makes the maximum operating frequency of each PE in the pulse array not strictly consistent

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
  • Convolutional neural network optimization method and device based on pulse array
  • Convolutional neural network optimization method and device based on pulse array
  • Convolutional neural network optimization method and device based on pulse array

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0041] In order to facilitate the understanding of the solutions provided by the embodiments of the present invention, CNN and pulse arrays are introduced first.

[0042] CNN usually consists of multiple convolutional layers (Convolutional Layer, Conv Layer). The two operational elements of convolution include input activation (InputActivation) matrix or input feature map (InputFeatureMap), and filter (Filter).

[0043] Input the activation matrix to preprocess the image data to form a matrix that conforms to the size of the CNN model; the filter is another set of matrices (called filter matrices) obtained through CNN model training, and the elements stored in the filter matrix It is called Weight.

[0044] The convolution operation performed by each convolutional layer consists of scanning over the input activation matrix with a filter. Among them, every time the filter scans to a position of the input activation matrix, it uses its weight to perform a matrix dot multiplicat...

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 provides a convolutional neural network optimization method and device based on a pulse array, so as to improve the lowest highest operation frequency in all PEs participating in calculation. The convolutional neural network comprises at least one convolutional layer. The convolution layer comprises at least one group of filters. The convolution operation executed by any convolutionlayer comprises the following steps: scanning on an input activation matrix by using a filter; wherein the pulse array is used for executing convolution calculation related to each convolution layer and comprises N rows and M columns of processing units PE; wherein the register of any PE participating in convolution calculation is used for storing the weight value of the corresponding filter. Themethod comprises the steps of determining an initial PE operation array for processing convolution operation of a convolution layer of the current layer; wherein the initial PE operation array comprises a row number n and a column number m; wherein n is smaller than or equal to N, and m is smaller than or equal to M; and replacing the PE with the lowest highest operation frequency in the initial PE operation array to obtain an optimized PE operation array.

Description

technical field [0001] The invention relates to the field of computers, in particular to a convolutional neural network optimization method and device based on pulse arrays. Background technique [0002] CNN (Convolution Neural Network, Convolution Neural Network) is one of the most popular image processing methods at this stage. Convolution is the largest and most time-consuming operation in the CNN calculation process. [0003] Pulse arrays can efficiently compute a large number of matrix dot products, which are the main computational mode of convolution operations. The pulse array is composed of a large number of processing units (ProcessingElement, PE), and all PEs must work at the same frequency. [0004] However, limited by the process technology, the physical parameters of many PEs cannot be strictly consistent, but are normally distributed, which makes the maximum operating frequency of each PE in the pulse array unable to be strictly consistent. Therefore, in ord...

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
IPC IPC(8): G06N3/04G06N3/063
CPCG06N3/049G06N3/063G06N3/045Y02D10/00
Inventor 谭婧炜佳马茂棣阎凯歌
Owner JILIN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products