Abstract:To make the problem of multilabel classification with many classes more tractable, in recent years, academia has seen efforts devoted to performing label space dimension reduction (LSDR). Specifically, LSDR encodes high-dimensional label vectors into low-dimensional code vectors lying in a latent space, so as to train predictive models at much lower costs. With respect to the prediction, it performs classification for any unseen instance by recovering a label vector from its predicted code vector via a decoding process. In this paper, we propose a novel method, namely End-to-End Feature-aware label space Encoding (E2FE), to perform LSDR. Instead of requiring an encoding function like most previous works, E2FE directly learns a code matrix formed by code vectors of the training instances in an end-to-end manner. Another distinct property of E2FE is its feature awareness attributable to the fact that the code matrix is learned by jointly maximizing the recoverability of the label space and the predictability of the latent space. Based on the learned code matrix, E2FE further trains predictive models to map instance features into code vectors, and also learns a linear decoding matrix for efficiently recovering the label vector of any unseen instance from its predicted code vector. Theoretical analyses show that both the code matrix and the linear decoding matrix in E2FE can be efficiently learned. Moreover, similar to previous works, E2FE can be specified to learn an encoding function. And it can also be extended with kernel tricks to handle nonlinear correlations between the feature space and the latent space. Comprehensive experiments conducted on diverse benchmark data sets with many classes show consistent performance gains of E2FE over the state-of-the-art methods.

Deep Cross-View Label Embedding with Correlation and Structure Preserved for Multi-Label Classification

Two-Stage Label Embedding Via Neural Factorization Machine for Multi-Label Classification

Learning and Exploiting Interclass Visual Correlations for Medical Image Classification

MULTI-LABEL IMAGE RECOGNITION WITH JOINT CLASS-AWARE MAP DISENTANGLING AND LABEL CORRELATION EMBEDDING

Disentangling, Embedding and Ranking Label Cues for Multi-Label Image Recognition

Multi-Label Classification with Label Graph Superimposing

Incorporating Label Embedding and Feature Augmentation for Multi-Dimensional Classification.

Embedding Regularizer Learning for Multi-View Semi-Supervised Classification.

Multi-View Embedding Learning for Incompletely Labeled Data.

End-to-End Feature-Aware Label Space Encoding for Multilabel Classification with Many Classes.

Label Embedding Based on Interclass Correlation Mining for Remote Sensing Image Scene Classification

A Label Embedding Method via Conditional Covariance Maximization for Multi-label Classification.

Masked Two-channel Decoupling Framework for Incomplete Multi-view Weak Multi-label Learning

Maximal Correlation Embedding Network for Multilabel Learning with Missing Labels

Deep Double Incomplete Multi-View Multi-Label Learning With Incomplete Labels and Missing Views

Multi-Label Learning by Exploiting Label Correlations with LDA

Dual-channel graph contrastive learning for multi-label classification with label-specific features and label correlations

Learning Deep Structure-Preserving Image-Text Embeddings

Co-attention network with label embedding for text classification

Cost-sensitive label embedding for multi-label classification

Label Embedding via Low-Coherence Matrices