Unsupervised transfer learning method based on graph convolution network

Abstract

The invention discloses an unsupervised transfer learning method based on a graph convolution network. The method comprises the steps of obtaining source domain and target domain samples for transferlearning from a database, performing feature extraction on the samples, and constructing a correlation graph; putting the sample features and the relational graph into a constructed graph convolutional network, mapping the sample features to a feature space with strong discrimination, and forming new features of the sample; performing distribution alignment on the learned feature space and the newsample features, so that the new sample features have good migration performance; meanwhile, constructing a classification network and learning a classifier of target domain data; and repeatedly using the gradient descent method until the loss functions of the graph convolution network and the classification network converge, and predicting the unlabeled target domain data. The method combines two characteristics of model discrimination capability and knowledge migration capability, can be used for a difficult unsupervised migration learning scene, and has good classification learning and data labeling capability in the application of an actual scene.

Description

technical field [0001] The invention relates to the field of unsupervised transfer learning, in particular to an unsupervised transfer learning method based on a graph convolutional network. Background technique [0002] In recent years, artificial intelligence technology has developed rapidly and played a strong role in various field application scenarios. The core of artificial intelligence technology is the learning and prediction of its algorithms. However, with the increase in the diversity and complexity of practical application scenarios. The shortcomings of traditional artificial intelligence algorithms are becoming more and more prominent. This is mainly due to the fact that traditional artificial intelligence algorithms assume that the learned data obeys the same distribution, and only by obeying such an assumption can they show good results. [0003] However, in practical applications, due to various reasons such as different times and different regions, the da...

AI Technical Summary

Problems solved by technology

[0003] However, in practical applications, due to various reasons such as different times and different regions, the data often do not completely obey the same distribution, which requires transfer learning technology to apply the learned knowledge from the source domain to the target domain.

A class of difficult and valuable practical problems is that there is a difference in the data distribution of the source domain and the target domain, and there is no labeled data in the target domain. This is the unsupervised transfer learning problem.

At present, there are some application methods that pay attention to this problem, but most of the methods to solve this problem have a shortcoming, that is, they ignore the geometric information of the data, which is of great significance for learning the correlation between samples.

Images