Research map / Learning theory and algorithms / Prediction and decision losses
Supervised learning: no loss no cry
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 |
|---|---|---|---|
| Replace SLIsotron with BregmanTron | Binary classification · Synthetic 2D Gaussian | AUC: 91.6% -> 92.3% | Helped |
| Replace SLIsotron with BregmanTron | Binary classification · MNIST 0 vs 8 | AUC: 94.6% -> 99.7% | Helped |
| Replace SLIsotron with BregmanTron | Binary classification · Fashion-MNIST odd vs even | AUC: 90.7% -> 97.9% | Helped |
| Use a feasible minimal-prediction solution instead of the global optimum in Step 4 (ablation) | BregmanTron · Synthetic 2D Gaussian | AUC: 92.3% -> 92.2% | Hurt |
| Use a feasible minimal-prediction solution instead of the global optimum in Step 4 (ablation) | BregmanTron · MNIST 0 vs 8 | AUC: 99.7% -> 99.3% | Hurt |
| Use a feasible minimal-prediction solution instead of the global optimum in Step 4 (ablation) | BregmanTron · Fashion-MNIST odd vs even | AUC: 97.9% -> 94.6% | Hurt |
| Fit the link against observed labels instead of the previous iteration's predictions (ablation) | BregmanTron · Synthetic 2D Gaussian | AUC: 92.3% -> 90.1% | Hurt |
| Fit the link against observed labels instead of the previous iteration's predictions (ablation) | BregmanTron · MNIST 0 vs 8 | AUC: 99.7% -> 99.6% | Hurt |
| Fit the link against observed labels instead of the previous iteration's predictions (ablation) | BregmanTron · Fashion-MNIST odd vs even | AUC: 97.9% -> 97.7% | Hurt |
| Use the loss learned on Fashion-MNIST 0 vs 6 to train a classifier on 2 vs 4 | Downstream classifier · Fashion-MNIST 2 vs 4 | AUC: 0.877 -> 0.879 | Helped |
| Compare BregmanTron with logistic regression | Binary classification · MNIST 0 vs 8 | AUC: 99.9% -> 99.7% | Hurt |
| Compare BregmanTron with logistic regression | Binary classification · Fashion-MNIST odd vs even | AUC: 98.5% -> 97.9% | Hurt |
Closest papers by meaning
- Learning with Fitzpatrick Losses (NeurIPS 2024 · similarity 0.86)
- Loss Functions and Operators Generated by f-Divergences (ICML 2025 · similarity 0.86)
- Realizable $H$-Consistent and Bayes-Consistent Loss Functions for Learning to Defer (NeurIPS 2024 · similarity 0.85)
- Multiclass Loss Geometry Matters for Generalization of Gradient Descent in Separable Classification (NeurIPS 2025 · similarity 0.85)
- Relating Misfit to Gain in Weak-to-Strong Generalization Beyond the Squared Loss (ICML 2025 · similarity 0.84)
- Newton Losses: Using Curvature Information for Learning with Differentiable Algorithms (NeurIPS 2024 · 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.