Research map / Label-efficient and robust learning / Recognition and long-tail
A Stitch in Time Saves Nine: A Train-Time Regularizing Loss for Improved Neural Network Calibration
Check 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 MDCA to focal loss (ablation) | ResNet32 · CIFAR10 | SCE (10^-3): 4.09 -> 3.22 | Helped |
| Add MDCA to focal loss (ablation) | ResNet32 · CIFAR10 | ECE (%): 1.01 -> 0.93 | Helped |
| Add MDCA to cross-entropy (ablation) | ResNet32 · CIFAR10 | SCE (10^-3): 8.68 -> 4.63 | Helped |
| Add MDCA to label smoothing (ablation) | ResNet32 · CIFAR10 | SCE (10^-3): 14.09 -> 12.38 | Helped |
| Add MDCA to focal loss (ablation) | ResNet20 · SVHN | SCE (10^-3): 2.54 -> 1.90 | Helped |
| Add MDCA to label smoothing (ablation) | ResNet20 · SVHN | SCE (10^-3): 18.80 -> 13.91 | Helped |
| Add MDCA to cross-entropy (ablation) | ResNet20 · SVHN | SCE (10^-3): 3.43 -> 1.46 | Helped |
| Increase training batch size away from 64 or 128 (ablation) | ResNet32 with MDCA · CIFAR10 | calibration performance | Hurt |
| Use a smaller or larger batch than 64 or 128 (ablation) | ResNet32 with MDCA · CIFAR10 | calibration performance | Hurt |
| Increase MDCA weight beta (ablation) | FL+MDCA · model-dataset pairs tested | SCE | Hurt |
| Add MDCA to focal loss (ablation) | ResNet18 · PACS domain shift average | SCE (10^-3): 12.06 -> 9.73 | Helped |
| Apply temperature scaling after FL+MDCA (ablation) | ResNet56 · CIFAR10 | SCE (10^-3): 2.93 -> 2.93 | No clear effect |
Closest papers by meaning
- A Benchmark Study on Calibration (ICLR 2024 · similarity 0.87)
- Towards Calibrated Multi-label Deep Neural Networks (CVPR 2024 · similarity 0.86)
- Uncertainty Weighted Gradients for Model Calibration (CVPR 2025 · similarity 0.86)
- Decoupling Feature Extraction and Classification Layers for Calibrated Neural Networks (ICML 2024 · similarity 0.85)
- Reassessing How to Compare and Improve the Calibration of Machine Learning Models (ICLR 2025 · similarity 0.85)
- Dirichlet-Based Prediction Calibration for Learning with Noisy Labels (AAAI 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.