Research map / Learning theory and algorithms / Prediction and decision losses
Two-Stage Learning to Defer with Multiple Experts
Anqi Mao, Christopher Mohri, Mehryar Mohri, Yutao Zhong
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 one expert (ResNet-10) to the base predictor | Two-stage surrogate loss (logistic) with ResNet-4 deferral · SVHN, no base cost | overall accuracy: 91.12 -> 91.85 ± 0.01% | Improved |
| go from single to two experts | Two-stage surrogate loss (logistic) with ResNet-4 deferral · SVHN, no base cost | overall accuracy: 91.85 ± 0.01% -> 92.77 ± 0.02% | Improved |
| go from two to three experts | Two-stage surrogate loss (logistic) with ResNet-4 deferral · SVHN, no base cost | overall accuracy: 92.77 ± 0.02% -> 93.30 ± 0.02% | Improved |
| add one expert to the base predictor | Two-stage surrogate loss (logistic) with ResNet-4 deferral · CIFAR-10, no base cost | overall accuracy: 70.56 -> 72.63 ± 0.20% | Improved |
| go from single to two experts | Two-stage surrogate loss (logistic) with ResNet-4 deferral · CIFAR-10, no base cost | overall accuracy: 72.63 ± 0.20% -> 75.84 ± 0.35% | Improved |
| go from two to three experts | Two-stage surrogate loss (logistic) with ResNet-4 deferral · CIFAR-10, no base cost | overall accuracy: 75.84 ± 0.35% -> 77.68 ± 0.07% | Improved |
| add one expert to the base predictor | Two-stage surrogate loss (logistic) with ResNet-4 deferral · SVHN, with base cost | overall accuracy: 91.12 -> 91.66 ± 0.01% | Improved |
| go from single to two experts | Two-stage surrogate loss (logistic) with ResNet-4 deferral · SVHN, with base cost | overall accuracy: 91.66 ± 0.01% -> 92.05 ± 0.10% | Improved |
| go from two to three experts | Two-stage surrogate loss (logistic) with ResNet-4 deferral · SVHN, with base cost | overall accuracy: 92.05 ± 0.10% -> 92.19 ± 0.03% | No clear effect |
| add one expert to the base predictor | Two-stage surrogate loss (logistic) with ResNet-4 deferral · CIFAR-10, with base cost | overall accuracy: 70.56 -> 71.73 ± 0.06% | Improved |
| go from single to two experts | Two-stage surrogate loss (logistic) with ResNet-4 deferral · CIFAR-10, with base cost | overall accuracy: 71.73 ± 0.06% -> 72.31 ± 0.31% | Improved |
| go from two to three experts | Two-stage surrogate loss (logistic) with ResNet-4 deferral · CIFAR-10, with base cost | overall accuracy: 72.31 ± 0.31% -> 72.42 ± 0.12% | No clear effect |
Closest papers by meaning
- Mastering Multiple-Expert Routing: Realizable $H$-Consistency and Strong Guarantees for Learning to Defer (ICML 2025 · similarity 0.92)
- Regression with Multi-Expert Deferral (ICML 2024 · similarity 0.92)
- A Two-Stage Learning-to-Defer Approach for Multi-Task Learning (ICML 2025 · similarity 0.90)
- Exploiting Human-AI Dependence for Learning to Defer (ICML 2024 · similarity 0.90)
- Realizable $H$-Consistent and Bayes-Consistent Loss Functions for Learning to Defer (NeurIPS 2024 · similarity 0.90)
- A Unifying Post-Processing Framework for Multi-Objective Learn-to-Defer Problems (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.
