Multi-label Learning With Non-equilibrium Labels Completions And Its Application

Multi-label learning is one of the main learning frameworks for dealing with the rich semantics of real-world objects.Multi-label learning is widely used in artificial intelligence and machine learning.In multi-label learning,each instance has multi-label attributes,which with local or global dependencies.Obviously,this relationship among labels can be used to obtain additional classification information,which helps to improve the performance of multi-label learning systems.In the real world,the number of labels in a sample is often much smaller than the number of unlabeled labels,otherwise the ambiguity of the example is meaningless.But it is undeniable that unknown labels may also contain a lot of valuable information.At present,many scholars consider that the correlation between labels is symmetric when considering the correlation learning algorithm,but the relationship among labels is not necessarily symmetrical.In view of the above problems,the main research works of this paper are as follows:(1)At present,many researchers usually directly add the label confidence matrix as a priori knowledge to the classification model,and do not consider the influence of non-equilibrium prior knowledge on the quality of the label set.Based on this,a multi-label learning algorithm is proposed,which uses kernel extreme learning machine with non-equilibrium label completion(KELM-NeLC).Firstly,information entropy is used to measure the correlation between labels which gets the basic label confidence matrix.Secondly,the basic label confidence matrix is improved to construct non-equilibrium label completion matrix by the non-equilibrium parameter.Finally,in order to learn obtain a more accurate label confidence matrix,the non-equilibrium label completion matrix is introduced with the kernel extreme learning machine to solve the multi-label classification problem.(2)We use the correlation of elements in the neighboring space to improve the quality of the neighboring labels space for the label correlation problem of the neighboring space.We propose a Non-equilibrium Labels Completion of Neighboring Labels Space algorithm(NeLC-NLS),which aims to improve the quality of the neighboring label space as to improve the performance of the multi-label classification.Firstly,the information entropy between labels is used to measure the strength of the relationship between labels,and the confidence matrix of the basic label is obtained.Then,the confidence matrix of non-equilibrium labels containing more information is obtained by using the proposed non-equilibrium label confidence matrix.Secondly,the similarity of samples is measured in the feature space and the k nearest neighbors are obtained.Then the non-equilibrium labels completion matrix isused to calculate the label completion matrix of the neighboring labels space.Finally,the extreme learning machine as a linear classifier is used to classify.(3)We consider the sample feature space information for reconstruction,and enhance the sample association of feature space while introducing labels correlations information.A mutil-label learning algorithm with Non-equilibrium Labels Completion and Mean Shift named NeLC-MS is proposed.The aim is to reconstruct the feature space and introduce labels correlation information to improve classifier classification performance.In the first step,the mean shift clustering method is used to reconstruct the information in the feature space.In the second step,information entropy is used to measure the correlation between labels which gets the basic label confidence matrix.Then the basic label confidence matrix is improved to construct a Non-equilibrium label completion matrix by the non-equilibrium parameter.Finally,the new training set is constructed by using the reconstructed feature space and the Non-equilibrium Labels Completion matrix,and the existing linear classifier is used for prediction based on the new training set.