Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Distance to Direction: Structure-aware Label-specific Feature Fusion for Label Distribution Learning

Suping Xu, Chuyi Dai, Lin Shang, Changbin Shao, Xibei Yang, Witold Pedrycz

TL;DR

This work tackles label ambiguity in LDL by extending prototype-based label-specific features with Structural Anchor Points (SAPs) to model cross-cluster interactions. It introduces LIFT-SAP, which augments distance and directional information relative to SAPs, and LDL-LIFT-SAP, a two-stage framework that fuses predictions from multiple LSF spaces into a cohesive label distribution. Through extensive experiments on 15 real-world LDL datasets, the authors demonstrate that LIFT-SAP outperforms LIFT, and LDL-LIFT-SAP surpasses seven strong baselines across diverse metrics. The approach highlights the value of structure-aware, multi-perspective feature representations for more accurate and robust LDL in practical applications.

Abstract

Label distribution learning (LDL) is an emerging learning paradigm designed to capture the relative importance of labels for each instance. Label-specific features (LSFs), constructed by LIFT, have proven effective for learning tasks with label ambiguity by leveraging clustering-based prototypes for each label to re-characterize instances. However, directly introducing LIFT into LDL tasks can be suboptimal, as the prototypes it collects primarily reflect intra-cluster relationships while neglecting cross-cluster interactions. Additionally, constructing LSFs using multi-perspective information, rather than relying solely on Euclidean distance, provides a more robust and comprehensive representation of instances, mitigating noise and bias that may arise from a single distance perspective. To address these limitations, we introduce Structural Anchor Points (SAPs) to capture inter-cluster interactions. This leads to a novel LSFs construction strategy, LIFT-SAP, which enhances LIFT by integrating both distance and directional information of each instance relative to SAPs. Furthermore, we propose a novel LDL algorithm, Label Distribution Learning via Label-specifIc FeaTure with SAPs (LDL-LIFT-SAP), which unifies multiple label description degrees predicted from different LSF spaces into a cohesive label distribution. Extensive experiments on 15 real-world datasets demonstrate the effectiveness of LIFT-SAP over LIFT, as well as the superiority of LDL-LIFT-SAP compared to seven other well-established algorithms.

From Distance to Direction: Structure-aware Label-specific Feature Fusion for Label Distribution Learning

TL;DR

This work tackles label ambiguity in LDL by extending prototype-based label-specific features with Structural Anchor Points (SAPs) to model cross-cluster interactions. It introduces LIFT-SAP, which augments distance and directional information relative to SAPs, and LDL-LIFT-SAP, a two-stage framework that fuses predictions from multiple LSF spaces into a cohesive label distribution. Through extensive experiments on 15 real-world LDL datasets, the authors demonstrate that LIFT-SAP outperforms LIFT, and LDL-LIFT-SAP surpasses seven strong baselines across diverse metrics. The approach highlights the value of structure-aware, multi-perspective feature representations for more accurate and robust LDL in practical applications.

Abstract

Label distribution learning (LDL) is an emerging learning paradigm designed to capture the relative importance of labels for each instance. Label-specific features (LSFs), constructed by LIFT, have proven effective for learning tasks with label ambiguity by leveraging clustering-based prototypes for each label to re-characterize instances. However, directly introducing LIFT into LDL tasks can be suboptimal, as the prototypes it collects primarily reflect intra-cluster relationships while neglecting cross-cluster interactions. Additionally, constructing LSFs using multi-perspective information, rather than relying solely on Euclidean distance, provides a more robust and comprehensive representation of instances, mitigating noise and bias that may arise from a single distance perspective. To address these limitations, we introduce Structural Anchor Points (SAPs) to capture inter-cluster interactions. This leads to a novel LSFs construction strategy, LIFT-SAP, which enhances LIFT by integrating both distance and directional information of each instance relative to SAPs. Furthermore, we propose a novel LDL algorithm, Label Distribution Learning via Label-specifIc FeaTure with SAPs (LDL-LIFT-SAP), which unifies multiple label description degrees predicted from different LSF spaces into a cohesive label distribution. Extensive experiments on 15 real-world datasets demonstrate the effectiveness of LIFT-SAP over LIFT, as well as the superiority of LDL-LIFT-SAP compared to seven other well-established algorithms.
Paper Structure (22 sections, 14 equations, 7 figures, 10 tables, 2 algorithms)

This paper contains 22 sections, 14 equations, 7 figures, 10 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Label Distribution Learning
  4. Label-specific Features Learning
  5. The LIFT-SAP Strategy
  6. Preliminaries
  7. LIFT for LDL
  8. Structural Anchor Point
  9. LSFs with SAPs
  10. Fusing LSFs with SAPs into LIFT
  11. The LDL-LIFT-SAP Algorithm
  12. Experiments
  13. Experimental Configuration
  14. Datasets
  15. Evaluation Metrics
  16. ...and 7 more sections

Figures (7)

  • Figure 1: An illustration of label distribution. (a) a landscape photo captured in Krattigen, Switzerland, containing five labels, i.e., 'Sky', 'Mountain', 'Lake', 'Grass', and 'House'. (b) The corresponding label distribution of the landscape photo, with description degrees of $0.32$, $0.21$, $0.28$, $0.13$, and $0.06$ for the above labels, respectively.
  • Figure 2: An illustration of label-specific features
  • Figure 3: Three types of LSFs constructed by (a) LIFT. (b) LIFT-SAP with distance information. (c) LIFT-SAP with directional information.
  • Figure 4: Label Distribution Learning via Label-specifIc FeaTure with Structural Anchor Points (LDL-LIFT-SAP)
  • Figure 5: CD diagrams on the six evaluation metrics.
  • ...and 2 more figures

Theorems & Definitions (1)

  • Remark 1