Research map / Architectures and efficiency / CNN design
The Reversible Residual Network: Backpropagation Without Storing Activations
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 reversible residual blocks instead of standard residual blocks | ResNet-32 / RevNet-38 · CIFAR-10 | classification error: 7.14% -> 7.24% | No clear effect |
| Use reversible residual blocks instead of standard residual blocks | ResNet-32 / RevNet-38 · CIFAR-100 | classification error: 29.95% -> 28.96% | Helped |
| Use reversible residual blocks instead of standard residual blocks | ResNet-110 / RevNet-110 · CIFAR-10 | classification error: 5.74% -> 5.76% | No clear effect |
| Use reversible residual blocks instead of standard residual blocks | ResNet-110 / RevNet-110 · CIFAR-100 | classification error: 26.44% -> 25.40% | Helped |
| Use reversible residual blocks instead of standard residual blocks | ResNet-164 / RevNet-164 · CIFAR-10 | classification error: 5.24% -> 5.17% | Helped |
| Use reversible residual blocks instead of standard residual blocks | ResNet-164 / RevNet-164 · CIFAR-100 | classification error: 23.37% -> 23.69% | No clear effect |
| Use reversible residual blocks instead of standard residual blocks | ResNet-101 / RevNet-104 · ImageNet single crop | Top-1 classification error: 23.01% -> 23.10% | No clear effect |
| Use channel-wise partitioning rather than columns, checkerboard, or rows (ablation) | RevNet · image classification experiments | classification performance | Helped |
| Reconstruct activations during backprop instead of using stored activations (ablation) | RevNet · CIFAR-10 | final performance | No clear effect |
| Use reversible residual stacks to reduce activation storage | ResNet-101 / RevNet · ImageNet | mini-batch size per GPU: 32 -> 128 | Helped |
Closest papers by meaning
- Simplified Reversible Residual Networks (NeurIPS 2026 · similarity 0.96)
- Invertible Residual Networks (ICML 2019 · similarity 0.90)
- Reversible Architectures for Arbitrarily Deep Residual Neural Networks (AAAI 2018 · similarity 0.90)
- Dr2Net: Dynamic Reversible Dual-Residual Networks for Memory-Efficient Finetuning (CVPR 2024 · similarity 0.89)
- Reversible Decoupling Network for Single Image Reflection Removal (CVPR 2025 · similarity 0.86)
- Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers (NeurIPS 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.