Heterogeneous View Feature Attention For Multi-label Classification

In real life,individual instances can be described from different perspectives,with multiple types of data features,while being associated with a set of labels.Since view information is not only consistent but also heterogeneous,the multi-view multi-label learning method is the basic framework to solve the problem,and it is difficult to efficiently utilize the information from different views.However,current researches often ignore the importance of view features in the process of feature extraction.Also,they do not take into account the non-consistency in the extraction of view common features for label prediction,which directly reduces the effectiveness of multi-view multi-label learning.For that,the research work in this paper are as follows.1)Considering the contribution of views,existing work often assigns weights to views based on the number of correctly predicted labels,regardless of the importance of the extracted features,which actually make a difference in the process of multi-label prediction.To address this problem,a multi-view multi-label learning framework with view feature attention allocation is proposed(MMFA): First,the common and private features between different views are obtained.Then,the original features are compared with the common features to obtain the similarity.Next,the attention weight matrix can be obtained by multiplying the similarity with the synergistic feature matrix.Finally,the acquired attention is used to reconstruct the synergistic feature matrix that indicates the semantic information of the view for multi-label prediction.A large number of comparative experiments and component analysis verify the performance of the algorithm.2)The extraction of shared features among multiple views for label prediction is a common multi-view multi-label learning method.However,previous approaches assumed that the number and association degree of shared features were the same across views.In fact,they differ in the number and degree of association.The above assumption can lead to a poor communicability of the views.Therefore,this paper proposes a multi-view multi-label learning method based on the inconsistent shared features extracted by the graph attention model(MMSFI-C): The first step is to extract the shared and private features of multiple views.Next,the graph attention mechanism is adopted to learn the association degree of shared features of different views and calculate the adjacency matrix and attention coefficient.The number of associations is determined by taking the obtained adjacency matrix as a mask matrix,while the association degree of shared features is measured by the attention weight matrix.Finally,the new shared features are obtained for multi-label prediction.We conducted experiments on seven datasets to compare MMSFI-C algorithm with seven advanced algorithms.The experimental results demonstrate the advantages of our algorithm.3)Multi-label classification of images can more accurately represent the information contained in the images.The existing Res Net model suffers from mislabeling and omission of labels during multi-label prediction.Therefore,this paper combines the core idea of MMFA model with the existing deep learning model Res Net,using the output feature maps of four bottleneck structures in Res Net as four views,and also using the feature attention allocation idea of MMFA model to prevent the network from losing important feature information during the training process.Experimental results show that the network incorporating MMFA solves the fault and omission problems of multi-label image classification to a certain extent in several realistic scenarios.This paper extracts common features and private features among multiple heterogeneous views by minimizing the adversarial loss,based on which multi-view multi-label classification is performed.The problem of inconsistent view feature importance is solved by the idea of view feature attention allocation.The problem of poor view communicability is solved by learning the weights among the common features of different views through the graph attention mechanism.Solving the misclassification problem by combining the Res Net image classification model with the MMFA algorithm.