Migration method and system based on deep residual error correction network

A residual error correction, network technology, applied in the field of transfer learning, can solve problems such as slow training speed, difficulty in convergence, and upper limit of JAN performance

Pending Publication Date: 2019-10-11
BEIJING INSTITUTE OF TECHNOLOGYGY
View PDF3 Cites 22 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

JAN uses adversarial thinking to train JMMD. Because adversarial learning is difficult to train, it will cause slow training speed and difficult convergence

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
  • Migration method and system based on deep residual error correction network
  • Migration method and system based on deep residual error correction network
  • Migration method and system based on deep residual error correction network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0090] According to the analysis of the prior art, the present invention proposes a transfer learning technique based on a deep residual correction network, by inserting a residual correction block (including multiple layers) into the deep network, and using the joint distribution of multiple layers Transfer feature-level and task-level knowledge from the source domain to the target domain.

[0091] Such as figure 2 Shown, the flow process of the present invention's proposed method is:

[0092] Step 1. Set the parameter values ​​in the pre-built target network model based on the source domain data set and the target domain data set in the network;

[0093] Step 2, based on the target network model after setting the parameter value, carry out image category classification to each data in the target domain data set, and obtain the corresponding category of each data;

[0094] Step 3, label the corresponding data in the target domain data set based on the category correspondin...

Embodiment 2

[0162] This embodiment is implemented using python, and the pytorch library is used to build a neural network, experiments are done on the Office-Home dataset, and the results are compared with other unsupervised transfer learning algorithms. Since there are four subsets of domains in the Office-Home dataset, if they are combined in pairs, there are 4*4=16 kinds of cross-domain classification tasks, such as the cross-domain classification operation from the source domain Ar to the target domain Cl, which is denoted as Ar- >Cl, other tasks are similar. Specifically include the following steps:

[0163] Step 1: Construct the training set and test set of samples in the source domain dataset and target domain dataset respectively, add random noise to each sample, and adjust the image size to 256×256 pixels. Then the source domain training data set and the target domain training data set are divided into several mini-batches with a sample size of 32.

[0164] Step 2: Construct th...

Embodiment 3

[0193] Based on the same inventive concept, this embodiment also provides a migration system based on a deep residual correction network, including:

[0194] A setting module is used to set the parameter values ​​in the pre-built target network model based on the source domain data set and the target domain data set in the network;

[0195] Prediction module, for based on the target network model after setting parameter value, carry out image category classification to each data in the target domain data set, obtain the category corresponding to each data;

[0196] The result module is used to label the corresponding data in the target domain data set based on the category corresponding to each data, and obtain the target domain data set with labels;

[0197] Wherein, the target network model is constructed based on a residual correction block and a loss function;

[0198] The source domain data set includes a plurality of pictures and labels corresponding to each picture; th...

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 migration method and system based on a deep residual error correction network. The method comprises the following steps: setting values of parameters in a pre-constructed target network model based on a source domain data set and a target domain data set in the network; based on the target network model with the set parameter values, carrying out image category classification on all the data in the target domain data set, and obtaining the category corresponding to each piece of data; labeling the corresponding data in the target domain data set based on the categorycorresponding to each piece of data to obtain a target domain data set with a label; wherein the target network model is constructed based on a residual error correction block and a loss function; wherein the source domain data set comprises a plurality of pictures and labels corresponding to the pictures; wherein the target domain data set comprises a plurality of pictures. According to the residual error correction block and the loss function provided by the concept of the invention, the generalization ability of the original network can be improved through deepening the network, so that thecross-domain image classification accuracy is improved.

Description

technical field [0001] The present invention relates to the field of migration learning, in particular to a migration method and system based on a deep residual correction network. Background technique [0002] In recent years, with the rapid growth of data scale and computing resources, machine learning has made great progress in both theory and practice, and has become one of the main technical foundations of big data analysis. In particular, the application of deep neural network DNNs in various learning tasks such as information retrieval, computer vision, and natural language processing has significantly improved the experimental performance. However, models that often perform well mainly rely on a large amount of labeled data, and labeling a sufficient amount of data is a time-consuming and expensive operation. Under the two requirements of annotation scarcity and data distribution heterogeneity, transfer learning is proposed, and its working principle is as follows: ...

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): G06K9/62G06N3/08
CPCG06N3/08G06F18/217G06F18/24
Inventor 李爽刘驰
Owner BEIJING INSTITUTE OF TECHNOLOGYGY
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
Try Eureka
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap