Research map / Architectures and efficiency / RNNs and normalization
Adaptive Neural Networks for Efficient Inference
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 learned early-exit policy instead of uniform exits | GoogLeNet · ImageNet 2012 validation, GoogLeNet@1 | speed-up: 2% -> 9% | Helped |
| Use learned early-exit policy instead of uniform exits | GoogLeNet · ImageNet 2012 validation, GoogLeNet@2 | speed-up: 9% -> 22% | Helped |
| Use learned early-exit policy instead of uniform exits | GoogLeNet · ImageNet 2012 validation, GoogLeNet@5 | speed-up: 20% -> 33% | Helped |
| Use learned early-exit policy instead of uniform exits | ResNet50 · ImageNet 2012 validation, ResNet50@1 | speed-up: 1% -> 8% | Helped |
| Use learned early-exit policy instead of uniform exits | ResNet50 · ImageNet 2012 validation, ResNet50@2 | speed-up: 12% -> 18% | Helped |
| Use learned early-exit policy instead of uniform exits | ResNet50 · ImageNet 2012 validation, ResNet50@5 | speed-up: 10% -> 22% | Helped |
| Route adaptively across AlexNet, GoogLeNet, and ResNet50 rather than always using ResNet50 | Adaptive network selection · ImageNet 2012 validation | speedup: 2.8x | Helped |
| Use prediction entropy as a policy feature | Adaptive early-exit and network-selection policies · ImageNet 2012 | policy performance | Helped |
| Route directly from AlexNet to ResNet50, excluding GoogLeNet feedback | Adaptive network selection · ImageNet 2012 validation | network-selection performance | Hurt |
| Use adaptive early exits within one network rather than selecting among different networks | Adaptive inference · ImageNet 2012 validation | speedup at accuracy trade-offs: 20-30% | Mixed |
Closest papers by meaning
- Adaptive Depth Networks with Skippable Sub-Paths (NeurIPS 2024 · similarity 0.88)
- Triple Wins: Boosting Accuracy, Robustness and Efficiency Together by Enabling Input-Adaptive Inference (ICLR 2020 · similarity 0.88)
- Shallow-Deep Networks: Understanding and Mitigating Network Overthinking (ICML 2019 · similarity 0.88)
- IO-Adam: Rethinking Memory-Efficient Adaptive Optimizers from Gradient Computation (ICML 2026 · similarity 0.88)
- Learning Anytime Predictions in Neural Networks via Adaptive Loss Balancing (AAAI 2019 · similarity 0.88)
- From Kernels to Features: A Multi-Scale Adaptive Theory of Feature Learning (ICML 2025 · 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.