Research map / Label-efficient and robust learning / Noisy labels
Training Deep Neural Networks via Direct Loss Minimization
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 |
|---|---|---|---|
| add 20% label noise, use pos-AP instead of hinge-AP | pos-AP · VOC2012 action classification | mean AP: 45.7 -> 57.8 | Helped |
| 30% label noise, pos-AP vs hinge-AP | pos-AP · VOC2012 action classification | mean AP: 30.0 -> 45.9 | Helped |
| 40% label noise, pos-AP vs hinge-AP | pos-AP · VOC2012 action classification | mean AP: 16.8 -> 29.6 | Helped |
| no label noise, pos-AP vs hinge-AP | pos-AP · VOC2012 action classification | mean AP: 69.9 vs 69.5 | No clear effect |
| 10% label noise, pos-AP vs x-ent | pos-AP · VOC2012 action classification | mean AP: 59.7 -> 62.5 | Helped |
| detection with no noise, pos-AP vs hinge-AP | pos-AP · VOC2012 detection | mean AP: 47.6 -> 48.5 | Helped |
| 20% label noise in detection, hinge-AP | hinge-AP · VOC2012 detection | mean AP: 47.6 -> 0.0 | Hurt |
| 20% label noise in detection, pos-AP | pos-AP · VOC2012 detection | mean AP: 48.5 -> 40.3 | Hurt |
| use negative update instead of positive update | direct loss minimization · synthetic AP task | AP | Hurt |
| use perceptron update | per-AP/per-01 · synthetic AP task | AP | Hurt |
| use 0-1 task loss instead of AP task loss | direct loss minimization · synthetic AP task | AP | Hurt |
| use batch size 2048 instead of 512 | pos-AP · VOC2012 detection | AP | Helped |
Closest papers by meaning
- Global Optimality in Neural Network Training (CVPR 2017 · similarity 0.87)
- A Convergence Theory for Deep Learning via Over-Parameterization (ICML 2019 · similarity 0.86)
- Penalizing Gradient Norm for Efficiently Improving Generalization in Deep Learning (ICML 2022 · similarity 0.86)
- Newton Losses: Using Curvature Information for Learning with Differentiable Algorithms (NeurIPS 2024 · similarity 0.86)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 2025 · similarity 0.85)
- Improving Generalization of Deep Neural Networks by Optimum Shifting (AAAI 2025 · 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.