Research map / Label-efficient and robust learning / Recognition and long-tail
Dynamic Programming for Instance Annotation in Multi-Instance Multi-Label Learning
Anh T. Pham, Raviv Raich, Xiaoli Z. Fern
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use exact dynamic-programming posterior (MLR) instead of max/softmax heuristics (Mfast, SIM) | instance annotation model · MSCV2 (inductive) | instance accuracy: 47.6 -> 54.8 vs Mfast; 46.7 -> 54.8 vs SIM | Improved |
| use exact DP posterior with union constraint instead of MCMC-based LSB | instance annotation model · Letter Carroll (inductive) | instance accuracy: 55.9 -> 67.7 | Improved |
| use MLR instead of LSB | instance annotation model · HJA bird (inductive) | instance accuracy: 66.3 -> 69.1 | Improved |
| use MLR instead of LSB | instance annotation model · Voc12 (inductive) | instance accuracy: 42.8 -> 43.2 | No clear effect |
| use MLR instead of LSB | instance annotation model · 50Salad (inductive) | instance accuracy: 76.8 -> 76.0 | No clear effect |
| transductive setting with MLR instead of Mfast/SIM/LSB | instance annotation model · MSCV2 (transductive) | instance accuracy: 70.6 -> 84.9 vs Mfast; 69.7 -> 84.9 vs SIM; 61.1 -> 84.9 vs LSB | Improved |
| tune baselines on bag-level metric instead of instance-level metric | Mfast · MSCV2 (inductive) | instance accuracy: 47.6 -> 41.5 | Hurt |
| tune baselines on bag-level metric instead of instance-level metric | Mfast · Letter Frost (inductive) | instance accuracy: 58.9 -> 45.4 | Hurt |
| violation of union assumption (HJA bird vs HJA union) | MLR · HJA | instance accuracy: 72.3 -> 69.1 | Hurt |
| add l2-norm regularization to MLR and SLR (ablation) | MLR · Voc12 (small training data) | instance accuracy | Improved |
| increase number of training bags | MLR · MSCV2 | instance accuracy: MLR matches SIM and LSB with 17% of their training data | Improved |
| increase training data from 10% to 90% | MLR · 50Salad | gap to SLR: 20% gap -> 1-2% gap | Improved |
Closest papers by meaning
- Discover Multiple Novel Labels in Multi-Instance Multi-Label Learning (AAAI 2017 · similarity 0.88)
- Multi-Instance Causal Representation Learning for Instance Label Prediction and Out-of-Distribution Generalization (NeurIPS 2022 · similarity 0.88)
- Fast Multi-Instance Multi-Label Learning (TPAMI 2018 · similarity 0.88)
- Multi-Instance Multi-Label Classification from Crowdsourced Labels (AAAI 2025 · similarity 0.86)
- Rethinking Learning from Label Proportions via Moment Matching (NeurIPS 2026 · similarity 0.85)
- Forming Auxiliary High-confident Instance-level Loss to Promote Learning from Label Proportions (CVPR 2025 · similarity 0.85)
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.
