Research map / Label-efficient and robust learning / Recognition and long-tail
A Primal Dual Formulation For Deep Learning With Constraints
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| add constrained learning (no inference constraints) | BiLSTM SRL baseline · SRL 1% data | F1: 62.99 -> 66.21 | Helped |
| add constrained learning on top of constrained inference | B+CI · SRL 1% data | F1: 67.90 -> 68.71 | Helped |
| add constrained learning on top of constrained inference | B+CI · SRL 10% data | F1: 78.63 -> 78.72 | Helped |
| add constrained learning | BiLSTM SRL baseline · SRL | constraint violations: 14,857 -> 9,406 | Helped |
| add constrained learning | TypeNet baseline · TypeNet 5% data | MAP: 68.62 -> 80.13 | Helped |
| add semi-supervised constrained learning (unlabeled data for constraints) | CL · TypeNet 5% data | MAP: 80.13 -> 82.22 | Helped |
| add constrained learning | TypeNet baseline · TypeNet 100% data | MAP: 70.47 -> 82.80 | Helped |
| add constrained learning | TypeNet baseline · TypeNet 5% data | constraint violations: 22,715 -> 25 | Helped |
| add hierarchical embedding loss H to baseline | B · TypeNet 5% data | MAP: 68.62 -> 68.71 | No clear effect |
| add constrained inference on top of constrained learning | CL/SCL · NER GMB | F1 | No clear effect |
| add semi-supervised constrained learning | CL · NER GMB | F1 | Helped |
| add constrained inference alone (early stopped dual decomposition) | baseline · NER GMB | constraint violations | Hurt |
Closest papers by meaning
- Gradient-Based Inference for Networks with Output Constraints (AAAI 2019 · similarity 0.86)
- On Regularization and Inference with Label Constraints (ICML 2023 · similarity 0.85)
- Constraining Linear-chain CRFs to Regular Languages (ICLR 2022 · similarity 0.85)
- Distilling Knowledge from Well-Informed Soft Labels for Neural Relation Extraction (AAAI 2020 · similarity 0.84)
- DL2: Training and Querying Neural Networks with Logic (ICML 2019 · similarity 0.84)
- Minimal Semantic Sufficiency Meets Unsupervised Domain Generalization (NeurIPS 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.