Research map / Learning theory and algorithms / Prediction and decision losses
The Lovász Hinge: A Novel Convex Surrogate for Submodular Losses
Jiaqian Yu, Matthew B. Blaschko
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 |
|---|---|---|---|
| train with Lovasz hinge instead of 0-1 SVM | submodular loss learning · synthetic early detection | test Delta1: 0.166 -> 0.100 | Improved |
| train with Lovasz hinge instead of slack rescaling | submodular loss learning · synthetic early detection | test Delta1: 0.144 -> 0.100 | Improved |
| train with Lovasz hinge instead of margin rescaling | submodular loss learning · synthetic early detection | test Delta1: 0.154 -> 0.100 | Improved |
| train with Lovasz hinge instead of 0-1 loss | submodular loss learning · synthetic early detection | test 0-1 loss: 2.87 -> 7.42 | Hurt |
| train with Lovasz hinge instead of 0-1 SVM | submodular loss learning · PASCAL VOC multilabel Delta2 | test Delta2: 0.5091 -> 0.4371 | Improved |
| train with Lovasz hinge instead of slack rescaling | submodular loss learning · PASCAL VOC multilabel Delta2 | test Delta2: 0.4927 -> 0.4371 | Improved |
| train with Lovasz hinge instead of margin rescaling | submodular loss learning · PASCAL VOC multilabel Delta2 | test Delta2: 0.4437 -> 0.4371 | Improved |
| train with Lovasz hinge instead of 0-1 SVM | submodular loss learning · PASCAL VOC multilabel Delta3 | test Delta3: 0.9877 -> 0.9447 | Improved |
| train with Lovasz hinge instead of margin rescaling | submodular loss learning · PASCAL VOC multilabel Delta3 | test Delta3: 0.9718 -> 0.9447 | Improved |
| train with Lovasz hinge instead of 0-1 SVM | submodular loss learning · PASCAL VOC multilabel Delta2 | test 0-1 loss: 0.8320 -> 0.9329 | Hurt |
| train with Lovasz hinge instead of margin rescaling | submodular loss learning · PASCAL VOC multilabel Delta2 | test 0-1 loss: 1.0010 -> 0.9329 | Improved |
Closest papers by meaning
- On Structured Prediction Theory with Calibrated Convex Surrogate Losses (NeurIPS 2017 · similarity 0.89)
- Establishing Linear Surrogate Regret Bounds for Convex Smooth Losses via Convolutional Fenchel–Young Losses (NeurIPS 2025 · similarity 0.88)
- Consistency Conditions for Differentiable Surrogate Losses (NeurIPS 2025 · similarity 0.88)
- Realizable $H$-Consistent and Bayes-Consistent Loss Functions for Learning to Defer (NeurIPS 2024 · similarity 0.87)
- Solving hidden monotone variational inequalities with surrogate losses (ICLR 2025 · similarity 0.87)
- Decision-Focused Learning with Directional Gradients (NeurIPS 2024 · similarity 0.86)
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.
