Research map / Learning theory and algorithms / Prediction and decision losses
Structured Prediction with Projection Oracles
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 |
|---|---|---|---|
| Use Birkhoff-polytope projection and decoding instead of squared loss with permutation decoding (ablation) | Label ranking · Label ranking datasets | Hamming loss: 5.70 -> 5.10 (Authorship); 7.11 -> 4.65 (Glass); 19.26 -> 2.96 (Iris); 9.04 -> 5.88 (Vehicle); 10.57 -> 8.76 (Vowel); 1. | Mixed |
| Use Birkhoff projection and decoding instead of unit cube projection and decoding (ablation) | Label ranking · Label ranking datasets | Hamming loss: 12.83 -> 5.10 (Authorship); 24.35 -> 4.65 (Glass); 27.78 -> 2.96 (Iris); 26.36 -> 5.88 (Vehicle); 43.71 -> 8.76 (Vowel) | Helped |
| Use unit cube or row-stochastic matrices instead of Birkhoff polytope for decoding (ablation) | Label ranking · Label ranking | Hamming loss | Hurt |
| Use Euclidean rather than KL projection (ablation) | Projection-based label-ranking losses · Label ranking | Hamming loss | No clear effect |
| Use order-simplex projection and decoding instead of ridge regression | Ordinal regression · 16 ordinal-regression datasets | Average MAE: 0.72 -> 0.43 | Helped |
| Add unit-cube projection to squared loss with order-simplex decoding (ablation) | Ordinal regression · 16 ordinal-regression datasets | Average MAE: 0.47 -> 0.45 | Helped |
| Replace unit-cube projection with order-simplex projection (ablation) | Ordinal regression · 16 ordinal-regression datasets | Average MAE: 0.45 -> 0.43 | Helped |
| Add order-simplex decoding to squared loss with rounding (ablation) | Ordinal regression · 16 ordinal-regression datasets | Average MAE: 0.72 -> 0.47 | Helped |
| Use knapsack-polytope projection and decoding instead of unit cube projection and decoding (ablation) | Multilabel classification · Birds | F1-score: 38.87 -> 39.43 | Helped |
| Use knapsack-polytope projection and decoding instead of unit cube projection and decoding (ablation) | Multilabel classification · Emotions | F1-score: 56.60 -> 62.57 | Helped |
| Use knapsack-polytope projection and decoding instead of unit cube projection and decoding (ablation) | Multilabel classification · Scene | F1-score: 61.06 -> 69.01 | Helped |
| Use knapsack-polytope projection and decoding instead of unit cube projection and decoding (ablation) | Multilabel classification · Cal500 | F1-score: 34.62 -> 34.61 | Hurt |
Closest papers by meaning
- On Structured Prediction Theory with Calibrated Convex Surrogate Losses (NeurIPS 2017 · similarity 0.89)
- Structured Prediction with Stronger Consistency Guarantees (NeurIPS 2023 · similarity 0.88)
- Multiclass Loss Geometry Matters for Generalization of Gradient Descent in Separable Classification (NeurIPS 2025 · similarity 0.86)
- Learning with Fitzpatrick Losses (NeurIPS 2024 · similarity 0.85)
- Trading off Consistency and Dimensionality of Convex Surrogates for Multiclass Classification (NeurIPS 2024 · similarity 0.85)
- Universal Online Convex Optimization with $1$ Projection per Round (NeurIPS 2024 · 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.