Multi-mark learning method

Abstract

The invention provides a training method for a multi-mark learning model. The method comprises the steps that a given multi-mark data training set is used to train the multi-mark learning model whichcomprises an automatic encoder and a multi-classification model; the loss function of the multi-mark learning model is examined, and the loss function of the multi-mark learning model comprises a lossterm of the automatic encoder and a loss term of the multi-classification model; if the loss function does not satisfy a predetermined accuracy requirement, the relevant parameters of the automatic encoder and the multi-classification model are iterated; and a training classifier corresponding to the multi-classification model and the optimization weight matrix and offset vector of the automaticencoder are acquired. According to the invention, an acquired training model multi-classifies data; a large amount of unmarked data can be effectively utilized; and the classification accuracy is improved.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a multi-label learning method. Background technique [0002] Multi-label learning refers to attaching multiple different class labels to data instances. For example, when classifying a piece of news, it may cover multiple topics such as economy and security; another example is to label a video, the content of the video may be Relevant to themes like sci-fi, comedy, action, etc. Specifically, suppose X=R d Represents the d-dimensional input data instance space, Y={y 1 ,y 2 ,...,y q} represents a label space with q labels, and the task of multi-label learning is to learn from multi-label data D={(x i ,Y i )|1≤i≤m} to learn the function h:X→Y. where, for each multilabel sample (x i ,Y i ), x i ∈X,x i is a d-dimensional vector (x i1 ,x i2 ,...,x id ) T , Y i is instance x i Has a collection of tags. After obtaining the multi-label function h, for a new in...

AI Technical Summary

Problems solved by technology

[0008] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide an improved multi-label learning method to improve the effect of multi-label learning when the labeled samples are insufficient and the feature dimension of the original data is too high

Images