Unlock instant, AI-driven research and patent intelligence for your innovation.

Image classification neural network architecture search method and device based on network clipping

A technology of neural network and search method, which is applied in the direction of neural architecture, neural learning method, biological neural network model, etc. It can solve the problems of reducing the size of the search space of derived architecture diversity and the inability to find models, etc., to achieve increased diversity and good performance Effect

Active Publication Date: 2020-12-04
ZHEJIANG UNIV +1
View PDF4 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, "local selection" greatly reduces the diversity of derived architectures and the size of the search space
DARTS cannot find a model that has two operations in the same set of candidate operations

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
  • Image classification neural network architecture search method and device based on network clipping
  • Image classification neural network architecture search method and device based on network clipping
  • Image classification neural network architecture search method and device based on network clipping

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0043] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0044] Such as figure 1 As shown, in a preferred embodiment of the present invention, a network cropping-based image classification neural network architecture search method (NAP for short) is provided, and its specific steps are as follows:

[0045] S1: A hyperparameterized network pre-built for image classification tasks, and a non-standardized architecture weight parameter is added to each candidate operation of the hyperparameterized network, and the architecture weight parameter is positively correlated with the importance of the corresponding candidate operation.

[0046] In the present invention, the hyperparameterized network of S1 needs to be constructed according to specific tasks, and its network construction form is similar to that of traditional DARTS, which is composed of repeated superposition of cells. In this embodiment, the s...

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 an image classification neural network architecture search method and device based on network clipping. The method comprises the following steps: firstly, constructing a super-network searched by a neural network architecture, and proposing to fit parameters of the network architecture by using a non-standardized intensity factor for learning; and then, carrying out networkclipping on the learned non-standardized intensity factor according to a provided standard to obtain an optimal network architecture. Compared with other methods, the method has the advantage that better performance can be achieved by using fewer parameters in an image classification task.

Description

technical field [0001] The invention relates to neural network architecture search, in particular to a neural network architecture search method based on network clipping on image classification tasks. Background technique [0002] In recent years, Neural Architecture Search (NAS, Neural Architecture Search), as a technology that can automatically involve neural network structures, has attracted more and more researchers' attention. The best architecture designed by NAS has achieved performance beyond the network architecture designed by humans in a variety of tasks, such as image classification, semantic segmentation, object detection, etc. Conventional NAS methods, including those based on reinforcement learning or evolutionary algorithms. These methods make difficult choices for some network architecture candidates, but they fall into a difficult problem, that is, they all require a lot of computing resources, prompting researchers to turn to gradient-based neural networ...

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): G06K9/62G06N3/04G06N3/08
CPCG06N3/082G06N3/045G06F18/241G06F18/214
Inventor 庄越挺汤斯亮肖俊丁亚东郁强蒋忆
Owner ZHEJIANG UNIV