Supercharge Your Innovation With Domain-Expert AI Agents!

Combined optimization method based on Gumbel-softmax technology

A combinatorial optimization and technology technology, applied in neural learning methods, genetic models, genetic laws, etc., can solve problems such as discrete variable space discontinuity, and achieve the effect of avoiding local optimization, improving diversity, and strong versatility

Pending Publication Date: 2020-09-08
BEIJING NORMAL UNIVERSITY +1
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Gradients are often only directly related to continuous variables, while discrete variables cannot be derived because of spatial discontinuity, so it is difficult to optimize using the gradient descent method

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
  • Combined optimization method based on Gumbel-softmax technology
  • Combined optimization method based on Gumbel-softmax technology
  • Combined optimization method based on Gumbel-softmax technology

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0046] The technical solution of the present invention will be described in detail below. Take modularity optimization on the karate club dataset as an example.

[0047] The basic process of the inventive method is as figure 1 , figure 2 As shown, it specifically includes the following steps:

[0048] Step 1. Define the problem

[0049] 1-1) Transform the problem into a form suitable for the framework. Input the adjacency matrix of the karate dataset, and calculate the objective function of modularity according to the formula:

[0050]

[0051] In the formula |ε| is the total number of edges in the graph, A ij is the adjacency matrix of the graph, k i and k j Indicates the degree of i-node and j-node, if i-node and j-node are classified into the same community, then δ(s i ,s j )=1, otherwise δ(s i ,s j )=0.

[0052] 1-2) Determine the input data of the problem. Input the karate club data set, that is, the adjacency matrix, the number of nodes is 34, the origin...

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 combined optimization method based on a Gumbel-softmax technology. The combined optimization method comprises the following steps of: obtaining a Gambel-softmax model; the application target of the method is a combined optimization problem, and the method is used for solving the problem of combination optimization. The method mainly aims at solving the problem of combinatorial optimization on a graph by using a deep learning technology, and relates to an automatic differential technology, a Gumbel-softmax sampling technology, an evolutionary strategy and a genetic algorithm in the field of deep learning. A combined optimization problem on a graph is converted into a differentiable continuous function optimization problem through a Gumbel-softmax re-parameterization technology, and then an automatic differential technology in the field of deep learning is used for optimizing parameters. By running on the GPU in parallel, the performance and efficiency of the algorithm can be effectively improved. According to the method, the optimal feasible solution of the combination optimization problem defined on the graph can be obtained, and the method can be used forthe network structure optimization problem.

Description

technical field [0001] The present invention relates to the field of computer technology, specifically belongs to the field of machine learning and optimization algorithms, and its application target is combinatorial optimization problems (combinatorial optimization problems). Automatic differentiation technique, Gumbel-softmax sampling technique, evolutionary strategy and genetic algorithm. Background technique [0002] There are many combinatorial optimization problems in our real life, such as production scheduling problem and traveling salesman problem. There are also a large number of combinatorial optimization problems defined on graphs in computer science, such as the maximum independent set (maximum independent set) and minimum vertex cover (minimal vertex cover) problems, etc.; there is also a spin glass model in the field of statistical physics, We need to find the optimal particle spin state combination on the specified graph structure, so that the ground state e...

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): G06Q10/04G06K9/62G06N20/00G06N3/08G06N3/12G06N3/00
CPCG06Q10/04G06N20/00G06N3/084G06N3/126G06N3/006G06F18/25
Inventor 李垚鑫张江
Owner BEIJING NORMAL UNIVERSITY
Features
  • R&D
  • Intellectual Property
  • Life Sciences
  • Materials
  • Tech Scout
Why Patsnap Eureka
  • Unparalleled Data Quality
  • Higher Quality Content
  • 60% Fewer Hallucinations
Social media
Patsnap Eureka Blog
Learn More

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

© 2025 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com