Research map / Label-efficient and robust learning / Domain adaptation and generalization
Confidence Estimation via Auxiliary Models
Charles Corbière, Nicolas Thome, Antoine Saporta, Tuan-Hung Vu, Matthieu Cord, Patrick Pérez
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 |
|---|---|---|---|
| add encoder fine-tuning to ConfidNet (ablation) | ConfidNet · MNIST (SmallConvNet) | AUPR (%): 44.54 -> 45.45 | Improved |
| add encoder fine-tuning to ConfidNet (ablation) | ConfidNet · CIFAR-100 (VGG16) | AUPR (%): 71.30 -> 73.55 | Improved |
| train ConfidNet on training set instead of validation set (ablation) | ConfidNet · MNIST (MLP) | AUPR (%): 38.22 -> 59.72 | Improved |
| train ConfidNet on training set instead of validation set (ablation) | ConfidNet · CIFAR-100 (VGG16) | AUPR (%): 73.82 -> 73.55 | No clear effect |
| use MSE loss on TCP instead of BCE (ablation) | ConfidNet · CIFAR-10 | AUPR (%): 47.95 -> 49.94 | Improved |
| use MSE loss on TCP instead of ranking loss (ablation) | ConfidNet · CIFAR-10 | AUPR (%): 44.04 -> 49.94 | Improved |
| use MSE loss instead of focal BCE (ablation) | ConfidNet · CIFAR-10 | AUPR (%): 47.76 -> 49.94 | Improved |
| use ConfidNet instead of MCP | MCP · CIFAR-100 (VGG16) | AUPR (%): 71.30 -> 73.55 | Improved |
| use ConfidNet instead of MC Dropout | MC Dropout · MNIST (SmallConvNet) | AUPR (%): 42.12 -> 45.45 | Improved |
| add adversarial loss to ConfidNet (ablation) | ConfidNet · GTA5->Cityscapes | mIoU (%): 47.6 -> 49.0 | Improved |
| add adversarial loss to multi-scale ConfidNet (ablation) | Multi-scale ConfidNet · GTA5->Cityscapes | mIoU (%): 48.5 -> 49.9 | Improved |
| add multi-scale architecture to ConfidNet (ablation) | ConfidNet · GTA5->Cityscapes | mIoU (%): 47.6 -> 48.5 | Improved |
Closest papers by meaning
- Addressing Failure Prediction by Learning Model Confidence (NeurIPS 2019 · similarity 0.94)
- Revisiting Confidence Estimation: Towards Reliable Failure Prediction (TPAMI 2023 · similarity 0.86)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.86)
- Confidence-Aware Learning for Deep Neural Networks (ICML 2020 · similarity 0.85)
- Random-Set Neural Networks (ICLR 2025 · similarity 0.84)
- Deep Probabilistic Supervision for Image Classification (NeurIPS 2026 · similarity 0.83)
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.
