Research map / Label-efficient and robust learning / Recognition and long-tail
When Does Confidence-Based Cascade Deferral Suffice?
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 |
|---|---|---|---|
| Use Relative Confidence (query model 2 confidence) instead of confidence-only deferral | confidence-based deferral · ImageNet dog specialist | accuracy vs deferral rate | Helped |
| Relative Confidence when model 2 is a generalist (100% non-dog data) | confidence-based deferral · ImageNet generalist | accuracy vs deferral rate | No clear effect |
| Post-hoc deferral (MaxProb, Diff-01, Diff-Prob) vs Confidence | confidence-based deferral · ImageNet dog specialist | accuracy vs deferral rate | Helped |
| Post-hoc deferral vs Confidence under synthetic label noise | confidence-based deferral · CIFAR-100 label noise | accuracy vs deferral rate | Helped |
| Diff-01 post-hoc rule vs Confidence on real label noise | confidence-based deferral · Mini-ImageNet 60% noise | accuracy vs deferral rate | Helped |
| Post-hoc deferral when no distribution shift (h=100) | confidence-based deferral · CIFAR-100 long-tail | accuracy vs deferral rate | No clear effect |
| Post-hoc deferral (Diff-01) as head classes decrease to 50 and 25 | confidence-based deferral · CIFAR-100 long-tail | accuracy vs deferral rate | Helped |
| Post-hoc deferral when model 2 is highly accurate | confidence-based deferral · MNIST | accuracy vs deferral rate | No clear effect |
| Post-hoc Diff-01 with 8% non-dog data for model 2 | post-hoc deferral · ImageNet dog specialist | train vs test accuracy | Hurt |
Closest papers by meaning
- Regression with Multi-Expert Deferral (ICML 2024 · similarity 0.85)
- Density-Ratio Losses for Post-Hoc Learning to Defer (NeurIPS 2026 · similarity 0.85)
- Gatekeeper: Improving Model Cascades Through Confidence Tuning (NeurIPS 2025 · similarity 0.84)
- Exploiting Human-AI Dependence for Learning to Defer (ICML 2024 · similarity 0.83)
- Confidence Difference Reflects Various Supervised Signals in Confidence-Difference Classification (ICML 2025 · similarity 0.82)
- Mitigating Label Bias in Machine Learning: Fairness through Confident Learning (AAAI 2024 · 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.