Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
A Learning Error Analysis for Structured Prediction with Approximate Inference
Yuanbin Wu, Man Lan, Shiliang Sun, Qi Zhang, Xuanjing Huang
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 |
|---|---|---|---|
| margin with rho tuned on dev set instead of vanilla structural SVM | structural SVM · 20 newsgroups (toy kth-best inference) | error rate | Improved |
| larger k (looser approximate inference) with rho margin | structural SVM with m_rho · 20 newsgroups | error rate improvement | Improved |
| rho<1 instead of rho=1 with greedy decoding (u=0) (ablation) | structured perceptron · CoNLL 2000 POS and SIGHAN cws | label accuracy: p < 0.01 | Improved |
| rho<1 with mixed Viterbi/gid decoding (0<u<1) (ablation) | structured perceptron · chunking, POS, cws | label accuracy: 0.2% on chu, 0.6% on pos and 2% on cws | Improved |
| rho<1 with exact Viterbi (u=1) (ablation) | structured perceptron · POS tagging | label accuracy | Improved |
| rho>1 instead of rho<1 (ablation) | structured perceptron · sequential labelling | label accuracy | Hurt |
| rho=1-10^-4 vs rho=1 (ablation) | second-order PA parser · Basque | UAS: 79.8 -> 80.3 | Improved |
| rho=1+10^-4 vs rho=1 (ablation) | second-order PA parser · Greek | UAS: 81.7 -> 82.5 | Improved |
| tune rho near 1 vs rho=1 (ablation) | second-order PA parser · Czech, Hungarian, Turkish | UAS: 82.8 -> 83.0 | Mixed |
| rho=1+10^-3 vs rho=1 (ablation) | second-order PA parser · Basque | UAS: 79.8 -> 79.4 | Hurt |
| second-order features vs first-order model | MSTParser PA · Basque | UAS: 79.4 -> 79.8 | Improved |
Closest papers by meaning
- Learning Approximate Inference Networks for Structured Prediction (ICLR 2018 · similarity 0.87)
- A Smoother Way to Train Structured Prediction Models (NeurIPS 2018 · similarity 0.85)
- Bandit and Delayed Feedback in Online Structured Prediction (NeurIPS 2025 · similarity 0.84)
- Learning to Reject with a Fixed Predictor: Application to Decontextualization (ICLR 2024 · similarity 0.83)
- An Information-Theoretic Analysis of In-Context Learning (ICML 2024 · similarity 0.83)
- Prediction Under Imperfect Compression: A Theory of Approximate MDL (NeurIPS 2026 · 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.
