Abstract:Semi-supervised graph based sparse learning investigates both label related information and inherent characteristic in data when only a small number of labels are available. How to fully explore the structure information in data space and feature space and well utilize the label information in the partial labeled data is the crux to improve the performance of a learning model. In this study, the latent representation learning is introduced into data space and feature space (dual space) simultaneously in the framework of the semi-supervised graph lased sparse learning. A novel semi-supervised feature selection algorithm called semi-supervised feature selection based on pairwise constraint-guided dual space latent representation learning and double sparse graphs discriminant (PCDLRD) is proposed. Latent representation learning is firstly used in semi-supervised feature selection considering dual graph information, which aims to capture the complex internal correlation structure information in data space and feature space and well guide the process of feature selection. Furthermore, the latent representation learning with the pairwise constraints is defined, which embeds the pairwise constraints into the latent representation learning in data space so as to fully exploit the intrinsic structure of the data in the help of some supervised information. A double sparse graph discriminative model is also integrated into the framework to enhance the discriminative ability of the method, which uses both labeled and unlabeled data. In addition, a robust dual space latent representation learning model based on L 2,1 − n o r m is established, which is more robust to outliers than the squared norm. A unified solution for the objective function is investigated and its convergence is proved theoretically and experimentally. Experimental results on twelve public datasets validate the viability of the proposed method and the effectiveness of the proposed method in classification tasks compared to the other six feature selection methods.

Discriminative Semi-Supervised Feature Selection Via a Class-Credible Pseudo-Label Learning Framework

Pseudo-Label Guided Structural Discriminative Subspace Learning for Unsupervised Feature Selection.

Learning Accurate Label-Specific Features From Partially Multilabeled Data

Partial Label Feature Selection: an Adaptive Approach

Adaptive Semi-Supervised Learning with Discriminative Least Squares Regression

Prototype-Based Pseudo-Label Refinement for Semi-Supervised Hyperspectral Image Classification

Confidence-Driven Semi-Supervised Partial Label Learning

A Semi-supervised Method for Feature Selection

Pseudo-label Selection for Deep Semi-supervised Learning

On Discriminative Semi-Supervised Classification.

Semi-supervised feature selection based on pairwise constraint-guided dual space latent representation learning and double sparse graphs discriminant

Dual-Decoupling Learning and Metric-Adaptive Thresholding for Semi-Supervised Multi-Label Learning

Self Adaptive Threshold Pseudo-labeling and Unreliable Sample Contrastive Loss for Semi-supervised Image Classification

Semi-supervised Local Feature Selection for Data Classification

Research on Multi-Label Semi-Supervised Learning Algorithm Based on Dual Selection Criteria

Prototype-Guided Pseudo Labeling for Semi-Supervised Text Classification

Semi-supervised Discriminant Analysis Based on Sparse-Coding Theory

Can Label-Specific Features Help Partial-Label Learning?

Confidence-Induced Granular Partial Label Feature Selection Via Dependency and Similarity

A Semi-Supervised Medical Image Classification Method Based on Combined Pseudo-Labeling and Distance Metric Consistency

Probabilistic Labeled Semi-supervised SVM