Classification model establishment method based on dynamic joint distribution alignment and application thereof

Abstract

The invention discloses a classification model establishment method based on dynamic joint distribution alignment and application thereof, and belongs to the field of domain adaptation, and the method comprises the following steps: weighting samples in a source domain data set and a target domain data set respectively, and enabling category distribution to be the same; inputting samples in the two data sets into a joint distribution alignment model, and calculating loss, wherein the joint distribution alignment model comprises: a feature extractor used for extracting features of an input sample; the feature discriminator that is used for judging the source of the features extracted by the feature extractor; the classifier that is used for classifying the features extracted by the feature extractor to generate corresponding category labels; and the category discriminator that is used for judging the source of the category label generated by the classifier; after updating the joint distribution alignment model according to the loss, labeling a pseudo label for a sample in the target domain data set to update a sample weight; after iterative training is finished, forming a classification model by a feature extractor and a classifier. The problem of lack of training data can be solved, and training resources and time are reduced.

Description

technical field [0001] The invention belongs to the field of domain adaptation, and more specifically relates to a method for establishing a classification model based on dynamic joint distribution alignment and its application. Background technique [0002] The traditional classification problem requires that the data of the training set and the test set are independent and identically distributed, and the richness of the training set data is conducive to the training and performance improvement of the classifier. With the continuous development of network technology, we have entered the era of big data, and a large amount of information data is generated every day, so that the classifier can rely on these data to continuously train and update the model, which greatly improves the performance of the classifier. However, few of these data are well-labeled, which increases the difficulty of classifier training, and manual labeling of data is very labor-intensive, which brings...

AI Technical Summary

Problems solved by technology

[0007] Aiming at the defects and improvement needs of the prior art, the present invention provides a classification model building method based on dynamic joint distribution alignment and its application, the purpose of which is to solve the problem of lack of training data in image or text classification based on the method of domain adaptation problems while effectively reducing the required training resources and time

Images