Research map / Learning theory and algorithms / Prediction and decision losses
Moment Distributionally Robust Tree Structured Prediction
Yeshu Li, Danyal Saeed, Xinhua Zhang, Brian D Ziebart, Kevin Gimpel
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 |
|---|---|---|---|
| replace BiAF cross-entropy layer with Marginal DRO loss, 10 training samples | tree DRO · PTB m=10 | UAS: 93.48 -> 94.51 | Improved |
| replace BiAF layer with Marginal DRO loss, 10 samples | tree DRO · UD Dutch m=10 | UAS: 90.86 -> 92.41 | Improved |
| replace BiAF layer with Marginal DRO loss, 10 samples | tree DRO · CTB m=10 | UAS: 88.45 -> 89.19 | Improved |
| replace BiAF layer with Marginal DRO loss, 10 samples, low-resource Turkish | tree DRO · UD Turkish m=10 | UAS: 17.64 -> 24.85 | Improved |
| replace BiAF layer with Marginal DRO loss with 1000 samples | tree DRO · PTB m=1000 | UAS: 97.16 -> 97.12 | No clear effect |
| Game algorithm instead of BiAF with 1000 samples | tree DRO Game · UD Turkish m=1000 | UAS: 42.82 -> 36.30 | Hurt |
| use Stochastic or Game instead of Marginal for training cost | tree DRO · CPU time per step | time (s): 0.28 -> 2.72 | Hurt |
| Game-theoretic solver instead of Marginal | tree DRO · CPU time per step | time (s): 0.28 -> 7.25 | Hurt |
| use ADMM instead of Frank-Wolfe for arborescence projection (ablation) | projection oracle · random-point projection | objective | Improved |
| use ADMM instead of Frank-Wolfe for projection (ablation) | projection oracle · per-iteration cost | cost: about 8n times higher | Hurt |
| vary smoothing mu from tiny to moderate (ablation) | Marginal · UD Dutch m=100 | UAS | Improved |
| small lambda (ambiguity radius) (ablation) | Marginal · UD Dutch m=100 | UAS | Hurt |
Closest papers by meaning
- On Certified Generalization in Structured Prediction (NeurIPS 2023 · similarity 0.86)
- SGA: A Robust Algorithm for Partial Recovery of Tree-Structured Graphical Models with Noisy Samples (ICML 2021 · similarity 0.86)
- Near-Optimal Decision Trees in a SPLIT Second (ICML 2025 · similarity 0.85)
- BARK: A Fully Bayesian Tree Kernel for Black-box Optimization (ICML 2025 · similarity 0.84)
- Distribution Learning Meets Graph Structure Sampling (NeurIPS 2025 · similarity 0.84)
- Multistage Defer Trees for Hybrid Interpretability: If at First You Can't Succeed, Tree Again (NeurIPS 2026 · 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.
