Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Machine learning method based on difficult sample learning and neural network fusion

A neural network and sample learning technology, applied in neural learning methods, biological neural network models, integrated learning, etc., can solve problems such as difficult learning accuracy and difficult sample learning, and achieve accurate fusion results

Pending Publication Date: 2020-11-20
BEIJING DEEP AI INTELLIGENT TECH +1
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] At present, machine learning is usually carried out on all given data sets. There is no way to learn specifically for difficult samples and integrate learning results with all data sets. As a result, although the learning results are generally good, further improvement is required. It is difficult to learn precision
[0004] At present, there is no same idea as the technology proposed in the present invention in the existing learning technology of neural network, and the learning methods related to it are transfer learning and difficult sample learning

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
  • Machine learning method based on difficult sample learning and neural network fusion
  • Machine learning method based on difficult sample learning and neural network fusion
  • Machine learning method based on difficult sample learning and neural network fusion

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0036] When the method of the present invention is specifically implemented, the neural network fusion is not limited to that described in step 1 and step 2, that is, neural network B 1 Not necessarily based on hard sample training; neural network A 1 with neural network B 1 Can be two separate networks obtained by other means;

[0037] Further, during specific implementation, the number of networks can be increased to more than two, and the network structures participating in the fusion can also be different, for example: "In the incremental learning, the network learned on the newly added training samples is fused with the original network, Or the situation of merging the networks obtained by two different learning methods is also a specific implementation derived from the method of the present invention.

[0038] This embodiment describes the specific implementation of learning the twin U-shaped neural network (hereinafter referred to as the collaborative segmentation net...

Embodiment 2

[0051] It should be noted that image 3 It is only a kind of neural network combination method, and other methods of combining the output results of two neurons into one are also applicable to the method of the present invention, including the following sub-steps:

[0052] Step 4): On all original samples, only for the co-segmentation network Learn the weights of all new neurons in the network to form a trained collaborative segmentation network

[0053] Step 5): Obtain the co-segmentation network Its structure and co-segmentation network and co-segmentation network have the same structure, and its parameters are the collaborative segmentation network learned in step 4 The weight of the newly added neuron in will be connected by the newly added neuron in the collaborative segmentation network co-segmentation network The result of the weighted combination of the parameters of the corresponding neurons.

[0054] So far, from step 1) to step 5), the first difficul...

Embodiment 3

[0069] The method can also be applied to solve the incremental learning problem of the neural network. After the neural network is learned on the existing training samples, when the training samples increase, the new neural network can only be learned on the newly added samples, and then the new neural network can be obtained by using the The above neural network fusion method fuses the new neural network with the existing neural network to obtain a refined neural network, which can not only adapt to the new data, but also maintain the calculation accuracy of the original data.

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 relates to a machine learning method based on difficult sample learning and neural network fusion, and belongs to the technical field of artificial intelligence. According to the method,learning is carried out on all training sample sets and difficult training sample sets to obtain two neural networks, and then the two neural networks are fused to obtain a new neural network with better performance; the difficult sample learning and network fusion process can be repeatedly carried out for many times until the performance of the neural network cannot be improved, so that the difficult sample calculation precision is effectively improved on the premise of keeping the easy sample calculation precision; the method is also suitable for networks obtained in any other modes, for example, networks obtained by learning on newly added training samples are fused with an original network in incremental learning, or networks obtained by two different learning methods are fused. According to the method, two or more different neural networks are fused into a network with better performance, the overall calculation is more reliable, and the precision is higher.

Description

technical field [0001] The invention relates to a machine learning method based on fusion of difficult sample learning and neural network, belonging to the technical field of artificial intelligence. Background technique [0002] Neural network learning, especially deep network learning, usually requires big data to achieve good learning results. However, a large amount of data is generally better for learning, which does not mean that all data results are good. There may be a small number of difficult-to-learn samples that are submerged in a large amount of data and cannot be fully learned, resulting in that the overall learning effect cannot be further improved. By learning alone on a small number of difficult samples, the characteristics of these samples can be highlighted, and then integrated with the overall learning results, it should be able to effectively improve the learning effect of the neural network. In addition, this learning method can not only be used to sol...

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): G06N20/20G06N3/08
CPCG06N20/20G06N3/08
Inventor 刘峡壁许肖汉
Owner BEIJING DEEP AI INTELLIGENT TECH
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
Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com