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

A Self-Supervised Domain Adaptive Deep Learning Method Based on Consistency Training

A deep learning and consistency technology, applied in neural learning methods, biological neural network models, instruments, etc., can solve problems such as difficulty in obtaining models and difficulty in convergence

Active Publication Date: 2020-10-27
NAT INNOVATION INST OF DEFENSE TECH PLA ACAD OF MILITARY SCI
View PDF7 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved by the present invention is to optimize a pair of objective functions against each other during domain confrontation training, and the convergence of the training process is difficult, and it is difficult to obtain a suitable model

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
  • A Self-Supervised Domain Adaptive Deep Learning Method Based on Consistency Training
  • A Self-Supervised Domain Adaptive Deep Learning Method Based on Consistency Training
  • A Self-Supervised Domain Adaptive Deep Learning Method Based on Consistency Training

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0033] The present invention will be further described in detail below in conjunction with the attached drawings and examples of self-adaptive learning in the field of image classification, but this does not limit the protection scope of the present invention.

[0034] figure 1 A schematic diagram of the self-supervised domain adaptive deep learning training process based on consistency training in the embodiment of the present invention is given. Adaptive learning in the field of image classification mainly includes the following steps:

[0035] S1: Construct a multi-task learning deep neural network, including a parameter θ e The feature extraction network E of the parameter is θ m Image classification network M, and the parameter is θ p The image enhancement transform prediction network P;

[0036] The image enhancement transformation in S1 in this embodiment adopts an image rotation operation.

[0037] S2: Transform the source domain image x s and its category label ...

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 self-supervised field adaptive deep learning method based on consistency training. This method first constructs a set of data augmentation transformations, and defines a label for each transformation. Construct a classification task for the source domain samples and their corresponding category labels; apply the data enhancement transformation to the source domain and target domain samples, and construct a self-supervised learning task by minimizing the error of predicting the transformation category; for the source domain and target domain Domain samples, by minimizing the KL divergence (Kullback-Leibler Divergence) of the output of the transformed samples and the original samples on the classification task, construct a consistent training task; construct a multi-task learning network, and combine the classification, automatic The supervised learning and consistency training tasks are jointly trained. This method does not need to label the samples in the target domain, can effectively learn the feature representation of the target domain, and improve the performance of sample classification and recognition in the target domain. The present application also discloses a domain-adaptive deep learning readable storage medium, which also has the above beneficial effects.

Description

technical field [0001] The invention belongs to the field of new generation information technology, and specifically relates to a field-adaptive deep learning image classification method and a readable storage medium. Background technique [0002] Machine learning, especially deep learning models usually require a large number of labeled samples for supervised learning. For example, the classification and recognition of images, texts, etc. need to collect a large number of samples, and also need to label the corresponding category of each sample. After the model is trained on the labeled data, it is applied to the test data. Supervised learning is a very effective method when the test data has the same distribution as the training data. However, in practical applications, the distribution of test data and training data is usually different, which leads to a sharp decline in the performance of the model on the test data set. [0003] Domain adaptation is a kind of technical...

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 Patents(China)
IPC IPC(8): G06N3/08G06N3/04G06K9/62
CPCG06N3/08G06N3/045G06F18/2155G06F18/24
Inventor 许娇龙肖良朱琪聂一鸣
Owner NAT INNOVATION INST OF DEFENSE TECH PLA ACAD OF MILITARY SCI
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