Research map / Learning theory and algorithms / Prediction and decision losses
Post-hoc estimators for learning to defer to an expert
Harikrishna Narasimhan, Wittawat Jitkrittum, Aditya Krishna Menon, Ankit Singh Rawat, Sanjiv Kumar
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 |
|---|---|---|---|
| use post-hoc thresholding instead of Chow when many samples are deferred | linear base classifier · synthetic two-subgroup data | accuracy vs fraction deferred | Improved |
| use post-hoc training instead of CSS and OvA losses | linear classifier · synthetic data | accuracy vs fraction deferred | Improved |
| use post-hoc thresholding instead of post-hoc training | linear classifier · synthetic data | accuracy | Hurt |
| use post-hoc estimators instead of CSS and OvA | ResNet-8 base with ResNet-56 expert, MobileNet-v2 with EfficientNet-B0 · CIFAR-10, CIFAR-100, ImageNet specialist deferral | accuracy vs fraction deferred | Improved |
| use threshold correction instead of post-hoc training | base models with specialist experts · CIFAR-10, CIFAR-100, ImageNet | accuracy | No clear effect |
| use CSS and OvA losses instead of softmax cross-entropy at large deferral cost c0 | base model · learning to defer benchmarks | accuracy | Hurt |
| use post-hoc estimators instead of CSS and OvA | ResNet-8 to ResNet-56 cascade, MobileNet-v2 to EfficientNet · CIFAR-100, ImageNet adaptive inference | accuracy vs FLOPs | Improved |
| use post-hoc estimators instead of Chow | cascades · CIFAR-100, ImageNet adaptive inference | accuracy vs FLOPs | No clear effect |
Closest papers by meaning
- Density-Ratio Losses for Post-Hoc Learning to Defer (NeurIPS 2026 · similarity 0.90)
- Regression with Multi-Expert Deferral (ICML 2024 · similarity 0.88)
- Exploiting Human-AI Dependence for Learning to Defer (ICML 2024 · similarity 0.87)
- Mastering Multiple-Expert Routing: Realizable $H$-Consistency and Strong Guarantees for Learning to Defer (ICML 2025 · similarity 0.87)
- A Unifying Post-Processing Framework for Multi-Objective Learn-to-Defer Problems (NeurIPS 2024 · similarity 0.86)
- Post-Hoc Reversal: Are We Selecting Models Prematurely? (NeurIPS 2024 · similarity 0.86)
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.
