Research map / Architectures and efficiency / RNNs and normalization
Learning to Cascade: Confidence Calibration for Improving the Accuracy and Computational Cost of Cascade Inference Systems
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 |
|---|---|---|---|
| Train the fast model with a loss that accounts for both its predictions and the expensive model's predictions | MobileNetV2 → ResNet18 cascade · CIFAR-100 | MACs^casc: 435.91 \pm 23.80 -> 383.08 \pm 13.85 | Helped |
| Train the fast model with a loss that accounts for both models' predictions | MobileNetV2 → ResNet152 cascade · CIFAR-100 | MACs^casc: 3100.55 \pm 238.31 -> 2645.08 \pm 136.97 | Helped |
| Use cascade-aware training in a three-model cascade | MobileNetV2 → ResNet18 → ResNet152 · CIFAR-100 | MACs^casc: 1.91 \pm 0.22 -> 1.68 \pm 0.13 | Helped |
| Use cascade-aware training in a three-model cascade | MobileNetV2 → ResNet18 → ResNet152 · CIFAR-100 | Acc^casc: 77.48 \pm 0.27 -> 77.53 \pm 0.18 | Helped |
| Train the fast model with a cascade-aware loss | MobileNetV2 → ResNet152 cascade · ImageNet | MACs^casc: 8.36 \pm 0.65 -> 8.08 \pm 0.21 | Helped |
| Apply the cascade-aware loss to early-exit training | MSDNet model splitting · CIFAR-100 | accuracy-computational cost trade-off | Helped |
| Add the cascade-aware loss | AlexNet fast model · CIFAR-100 fast-model classification | accuracy: -0.30; -0.48 | Hurt |
| Add the cascade-aware loss | VGG11 fast model · CIFAR-100 fast-model classification | accuracy: -0.21 | Hurt |
| Add the cascade-aware loss | MobileNetV2 fast model · CIFAR-100 fast-model classification | accuracy: +0.10; +0.08 | Helped |
| Add the cascade-aware loss | MobileNetV2 fast model · ImageNet fast-model classification | accuracy: +0.28 | Helped |
| Use Temperature Scaling rather than the cascade-aware loss | Cascade inference systems · CIFAR-100 model cascading | accuracy-computational cost trade-off | Mixed |
| Use confidence calibration that targets fast-model correctness alone | Cascade inference systems · CIFAR-100 model cascading | accuracy-computational cost trade-off | Mixed |
Closest papers by meaning
- When Does Confidence-Based Cascade Deferral Suffice? (NeurIPS 2023 · similarity 0.88)
- Learning for Single-Shot Confidence Calibration in Deep Neural Networks Through Stochastic Inferences (CVPR 2019 · similarity 0.87)
- Model Cascades with Provable Per-Class Quality (NeurIPS 2026 · similarity 0.86)
- A Benchmark Study on Calibration (ICLR 2024 · similarity 0.86)
- Gatekeeper: Improving Model Cascades Through Confidence Tuning (NeurIPS 2025 · similarity 0.86)
- Reassessing How to Compare and Improve the Calibration of Machine Learning Models (ICLR 2025 · similarity 0.84)
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.