Research map / Label-efficient and robust learning / Recognition and long-tail
Reconsidering Feature Structure Information and Latent Space Alignment in Partial Multi-label Feature Selection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes PML-FSLA, a partial multi-label feature selection method that uses latent-space alignment between feature and label projections (with OPTICS-chosen latent dimension) to disambiguate labels and scores features by the product Q·R of their latent coefficient matrices.
Problem
Feature selection for partial multi-label data, where candidate label sets contain noisy labels; existing methods underuse feature-space information and the feature selection terms they use (linear W or latent coefficients) are not robust to redundancy, noise, and structural inconsistency, and they
Approach
Align latent projections of features and labels (noise is assumed to lie in the label space) to disambiguate labels, and use the product QR of feature and label latent coefficient matrices as the feature-selection term so that selected features reflect latent similarity to positive labels.
Key results
- PML-FSLA achieves the best Micro-F1 and Macro-F1 on all eight datasets, e.g., Micro-F1 0.538 on CAL vs 0.149 for the Q-only variant PML-FSLA-q.
- PML-FSLA ranks first in about 85% of all dataset-metric-percentage cases, but not on Average Precision, Ranking Loss and Coverage for CHD_49 and Water.
- Ablation shows the QR feature selection term beats the Q-only term by a wide margin on Micro-F1 and Macro-F1 across seven datasets.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use PML-FSLA instead of the strongest listed baseline | PML-FSLA · CAL, CHD_49, Chess, Corel5K, LLOG_F, Water, and Yeast | Micro-F1 | Helped |
| Use PML-FSLA instead of the strongest listed baseline | PML-FSLA · HumanPseAAC | Micro-F1 | No clear effect |
| Use QR rather than Q alone for feature selection (ablation) | PML-FSLA · CAL, CHD_49, Chess, Corel5K, LLOG_F, Water, and Yeast | Micro-F1: 0.149 vs 0.538; 0.477 vs 0.637; 0.000 vs 0.579; 0.000 vs 0.263; 0.022 vs 0.339; 0.222 vs 0.556; 0.121 vs 0.496 | Helped |
| Use QR rather than Q alone for feature selection (ablation) | PML-FSLA · CAL, CHD_49, Chess, Corel5K, LLOG_F, Water, and Yeast | Macro-F1: 0.148 vs 0.538; 0.472 vs 0.615; 0.000 vs 0.588; 0.000 vs 0.231; 0.020 vs 0.321; 0.212 vs 0.551; 0.132 vs 0.488 | Helped |
Surprising findings
- The paper reports that PML-FSLA is weaker than baselines on some Average Precision, Ranking Loss and Coverage results (e.g., CHD_49, Water), which it attributes to a trade-off favoring positive-label identification.
Limitations
- Joint objective is nonconvex and only a local optimum is obtained via multiplicative updates.
- Trade-off: gains in positive-label metrics come at a cost in Average Precision, Ranking Loss and Coverage on some datasets.
- Parameter sensitivity reported only on CHD_49 with Ranking Loss.
Experimental setup
Datasets: HumanPseAAC CAL500 Chess Corel5K LLOG_F Water Yeast CHD_49
Benchmarks: HumanPseAAC CAL500 Chess Corel5K LLOG_F Water Yeast CHD_49
Metrics: Micro-F1 Macro-F1 Average Precision Ranking Loss Coverage
Baselines: PML-LC PML-FP PAR-VLS PAR-MAP FPML PML-FSSO MLKNN MIFS DRMFS
Builds on: OPTICS DBSCAN partial multi-label learning feature selection with latent space
Closest papers by meaning
- Navigating Real-World Partial Label Learning: Unveiling Fine-Grained Images with Attributes (AAAI 2024 · similarity 0.86)
- Double-Layer Hybrid-Label Identification Feature Selection for Multi-View Multi-Label Learning (AAAI 2024 · similarity 0.85)
- Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 2024 · similarity 0.84)
- Addressing Multi-Label Learning with Partial Labels: From Sample Selection to Label Selection (AAAI 2025 · similarity 0.84)
- CFDM: Contrastive Fusion and Disambiguation for Multi-View Partial-Label Learning (AAAI 2025 · similarity 0.84)
- Disentangled Partial Label Learning (AAAI 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.