Research map / Label-efficient and robust learning / Test-time adaptation
Addressing Failure Prediction by Learning Model Confidence
Charles Corbière, Nicolas THOME, Avner Bar-Hen, 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 |
|---|---|---|---|
| learn TCP confidence with ConfidNet instead of using the strongest reported baseline | failure prediction models · MNIST (MLP) | AUPR-Error: 52.18 -> 57.37 | Improved |
| learn TCP confidence with ConfidNet instead of using the strongest reported baseline | failure prediction models · MNIST (Small ConvNet) | AUPR-Error: 38.50 -> 45.89 | Improved |
| learn TCP confidence with ConfidNet instead of using the strongest reported baseline | failure prediction models · SVHN (Small ConvNet) | AUPR-Error: 48.18 -> 50.72 | Improved |
| learn TCP confidence with ConfidNet instead of using the strongest reported baseline | failure prediction models · CIFAR-10 (VGG-16) | AUPR-Error: 46.40 -> 49.94 | Improved |
| learn TCP confidence with ConfidNet instead of using the strongest reported baseline | failure prediction models · CIFAR-100 (VGG-16) | AUPR-Error: 72.59 -> 73.68 | Improved |
| learn TCP confidence with ConfidNet instead of using the strongest reported baseline | failure prediction models · CamVid (SegNet) | AUPR-Error: 49.35 -> 50.51 | Improved |
| add ConvNet fine-tuning after confidence training (ablation) | ConfidNet · MNIST (Small ConvNet) | AUPR-Error: 43.94% -> 45.89% | Improved |
| add ConvNet fine-tuning after confidence training (ablation) | ConfidNet · CIFAR-100 (VGG-16) | AUPR-Error: 72.68% -> 73.68% | Improved |
| replace MSE with BCE loss (ablation) | ConfidNet · CIFAR-10 (VGG-16) | AUPR-Error: 49.94% -> 47.95% | Hurt |
| replace MSE with BCE loss (ablation) | ConfidNet · CamVid (SegNet) | AUPR-Error: 50.51% -> 48.96% | Hurt |
| replace TCP with normalized TCPᵣ criterion (ablation) | ConfidNet · CIFAR-10 (VGG-16) | AUPR-Error: 49.94% -> 48.78% | Hurt |
| replace TCP with normalized TCPᵣ criterion (ablation) | ConfidNet · CamVid (SegNet) | AUPR-Error: 50.51% -> 51.35% | Improved |
Closest papers by meaning
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.85)
- Interpretable Failure Detection with Human-Level Concepts (AAAI 2025 · similarity 0.84)
- Pointwise Information Measures as Confidence Estimators in Deep Neural Networks: A Comparative Study (ICML 2025 · similarity 0.83)
- Typicalness-Aware Learning for Failure Detection (NeurIPS 2024 · similarity 0.83)
- Random-Set Neural Networks (ICLR 2025 · similarity 0.83)
- Deep Probabilistic Supervision for Image Classification (NeurIPS 2026 · similarity 0.82)
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.
