Research map / Label-efficient and robust learning / Recognition and long-tail
Multi-Label Open Set Recognition
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.
Introduces multi-label open set recognition (MLOSR) and SLAN, a non-deep method that uses sparse-reconstruction structure in feature space to create sub-labeling information, jointly learning a multi-label classifier and an open-set recognizer.
Problem
Multi-label classification where test instances may carry unknown labels that never appeared in training, requiring both correct prediction of known labels and recognition of unknown ones.
Approach
Builds a sparse reconstruction graph over training instances and uses it to enrich per-label sub-labeling information; an instance is flagged as having unknown labels when its sub-label representation departs from the holistic labeling information, and the classifier and recognizer are optimized jointly via alternating optimization.
Key results
- SLAN achieves superior or comparable performance to compared multi-label methods in 91.7% of cases across metrics (per the paper's summary of Table 6 win/tie/loss counts).
- On bibtex with 0 unknown-label batch, SLAN's ranking loss is 0.114 versus 0.217 for LIFT and 0.396 for MUENLPLR.
- SLAN beats anomaly detection baselines OC-SVM and IFOREST on F-measure in 26 of 30 dataset-setting comparisons (win/tie/loss 26/4/0).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use SLAN instead of the compared multi-label methods | Multi-label classification in MLOSR · Eight datasets, open-set label batches | Multi-label evaluation metrics | Mixed |
| Use SLAN instead of anomaly-detection baselines | Unknown-label recognition · Eight datasets, open-set label batches | F-measure | Helped |
Surprising findings
- Multi-class anomaly detection methods tend to reject instances with unknown labels because unknown labels co-occur with known labels in dense regions of feature space.
Limitations
- SLAN enriches #labels+1 sub-labeling information, so it hardly generalizes to extreme multi-label datasets.
- Operates on shallow multi-label feature representations, which may be less informative; no deep MLOSR model.
- Evaluation metrics for MLOSR are not yet extended.
Experimental setup
Datasets: llog enron slashdot recreation corel5k arts education rcvsubset2-2 bibtex
Benchmarks: llog enron slashdot recreation corel5k arts education rcvsubset2-2 bibtex
Metrics: ranking loss one-error coverage average precision macro-averaging AUC F-measure
Baselines: LIFT MUENLPLR SENCE LIMIC OC-SVM IFOREST MUENLFOREST
Builds on: open set recognition (1-vs-Set machine) streaming multi-label learning manifold assumption label enrichment ADMM
Closest papers by meaning
- Classifier-induced Reciprocal Points for Multi-label Open-set Recognition (AAAI 2026 · similarity 0.89)
- Unlocking the Power of Open Set: A New Perspective for Open-Set Noisy Label Learning (AAAI 2024 · similarity 0.85)
- From Coarse to Fine-Grained Open-Set Recognition (CVPR 2024 · similarity 0.84)
- Addressing Multi-Label Learning with Partial Labels: From Sample Selection to Label Selection (AAAI 2025 · similarity 0.83)
- Double-Layer Hybrid-Label Identification Feature Selection for Multi-View Multi-Label Learning (AAAI 2024 · similarity 0.83)
- ML-GOOD: Towards Multi-Label Graph Out-Of-Distribution Detection (AAAI 2025 · similarity 0.83)
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.