Research map / Architectures and efficiency / CNN design
Neural Routing by Memory
Kaipeng Zhang, Zhenqiang Li, Zhifeng Li, Wei Liu, Yoichi Sato
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 Conditional Attention module (ablation) | RMN (ResNet-18) · Tiny ImageNet | accuracy: 64.3% -> 66.5% | Improved |
| add Routing by Memory (RMN) to VGG-16 | VGG-16 · Tiny ImageNet | accuracy: 63.16% -> 66.57% | Improved |
| add Routing by Memory (RMN) to ResNet-18 | ResNet-18 · Tiny ImageNet | accuracy: 61.78% -> 66.53% | Improved |
| add Routing by Memory (RMN) to ResNet-50 | ResNet-50 · ImageNet | accuracy: 76.08% -> 78.31% | Improved |
| add Routing by Memory (RMN) to EfficientNet-B0 | EfficientNet-B0 · Tiny ImageNet | accuracy: 67.14% -> 68.85% | Improved |
| add Routing by Memory (RMN) to EfficientNet-B0 | EfficientNet-B0 · CIFAR-100 | accuracy: 78.04% -> 78.88% | Improved |
| add Routing by Memory (RMN) to VGG-16 | VGG-16 · ImageNet | accuracy: 72.92% -> 76.16% | Improved |
| RMN ResNet-18 vs original ResNet-50 | RMN (ResNet-18) · CIFAR-100 | accuracy: 77.53% -> 77.94% | Improved |
| replace RNR with RMN (N=4, no CA) | RMN (ResNet-18) · CIFAR-100 | accuracy: 78.42% -> 78.96% | Improved |
| replace RNR with RMN (N=4, no CA) | RMN (ResNet-18) · CIFAR-100 | latency per image: 59.8ms -> 14.2ms | Improved |
| replace DeepMoE with RMN (N=8, no CA) | ResNet-50 · ImageNet | accuracy gain over own baseline: 76.15% -> 77.12% vs 76.08% -> 77.78% | Improved |
| use too small batch size (128 or 256) for RMN (ablation) | RMN (ResNet-18) · Tiny ImageNet | accuracy | Hurt |
Closest papers by meaning
- Runtime Network Routing for Efficient Image Classification (TPAMI 2018 · similarity 0.86)
- Exploring Hidden Dimensions in Accelerating Convolutional Neural Networks (ICML 2018 · similarity 0.85)
- Multigrid Neural Memory (ICML 2020 · similarity 0.85)
- Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks (CVPR 2023 · similarity 0.85)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.84)
- Towards flexible perception with visual memory (ICML 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.
