Research map / Architectures and efficiency / RNNs and normalization
Zero Time Waste: Recycling Predictions in Early Exit Neural Networks
Maciej Wolczyk, Bartosz Wójcik, Klaudia Bałazy, Igor T. Podolak, Jacek Tabor, Marek Śmieja, Tomasz Trzcinski
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 |
|---|---|---|---|
| Reuse previous internal-classifier outputs with cascade connections and geometric ensembling | Early-exit ResNet-50 · ImageNet (25% compute) | accuracy: 33.8 -> 34.9 | Improved |
| Reuse previous internal-classifier outputs with cascade connections and geometric ensembling | Early-exit ResNet-50 · ImageNet (50% compute) | accuracy: 53.8 -> 54.9 | Improved |
| Reuse previous internal-classifier outputs with cascade connections and geometric ensembling | Early-exit ResNet-50 · OCT2017 (25% compute) | accuracy: 81.5 -> 89.4 | Improved |
| Use both cascade connections and geometric ensembling rather than either technique alone (ablation) | ZTW · CIFAR-100 on ResNet-56 and VGG-16 | accuracy | Improved |
| Use geometric rather than additive ensembling (ablation) | ZTW with cascade connections · ResNet-56 on CIFAR-10, CIFAR-100, and Tiny ImageNet | accuracy | Improved |
| Stop gradients through cascade connections (ablation) | ZTW · ResNet-56 on CIFAR-10, CIFAR-100, and Tiny ImageNet | accuracy | Mixed |
| Apply ZTW to binary rather than 10-class classification (ablation) | Early-exit models · CIFAR-2 | accuracy | Mixed |
Closest papers by meaning
- Fast yet Safe: Early-Exiting with Risk Control (NeurIPS 2024 · similarity 0.83)
- Beyond Greedy Exits: Improved Early Exit Decisions for Risk Control and Reliability (NeurIPS 2025 · similarity 0.83)
- BEEM: Boosting Performance of Early Exit DNNs using Multi-Exit Classifiers as Experts (ICLR 2025 · similarity 0.83)
- Jointly-Learned Exit and Inference for a Dynamic Neural Network (ICLR 2024 · similarity 0.82)
- DeepZero: Scaling Up Zeroth-Order Optimization for Deep Model Training (ICLR 2024 · similarity 0.82)
- How to Train Your Multi-Exit Model? Analyzing the Impact of Training Strategies (ICML 2025 · 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.
