Research map / Architectures and efficiency / RNNs and normalization
RDI-Net: Relational Dynamic Inference Networks
Huanyu Wang, Songyuan Li, Shihao Su, Zequn Qin, Xi Li
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 Uniform Sampling Warm-up (ablation) | RDI-Net · CIFAR-10 (ResNet-110) | accuracy (%): 93.21 -> 93.52 | Improved |
| add Uniform Sampling Warm-up (ablation) | RDI-Net · CIFAR-100 (ResNet-110) | accuracy (%): 70.32 -> 70.81 | Improved |
| add relational routers on top of USW (ablation) | RDI-Net · CIFAR-10 (ResNet-110) | accuracy (%): 93.52 -> 94.57 | Improved |
| add relational routers on top of USW (ablation) | RDI-Net · CIFAR-100 (ResNet-110) | accuracy (%): 70.81 -> 73.71 | Improved |
| add Sample Relation Module (ablation) | RDI-Net · CIFAR-10 (ResNet-110) | accuracy (%): 94.57 -> 95.06 | Improved |
| add Sample Relation Module (ablation) | RDI-Net · CIFAR-10 (ResNet-110) | GFLOPs: 0.31 -> 0.25 | Improved |
| add Sample Relation Module (ablation) | RDI-Net · CIFAR-100 (ResNet-110) | accuracy (%): 73.71 -> 74.11 | Improved |
| use decay rate gamma 0.5 instead of 0 (no relation between routers) (ablation) | RDI-Net · CIFAR-10 (ResNet-110) | accuracy (%): 93.21 -> 94.87 | Improved |
| use decay rate gamma 0.5 instead of gamma 1.0 (sum of features) (ablation) | RDI-Net · CIFAR-10 (ResNet-110) | accuracy (%): 93.81 -> 94.87 | Improved |
| use learnable edge weights instead of fixed gamma 0.5 (ablation) | RDI-Net · CIFAR-10 (ResNet-110) | accuracy (%): 94.87 -> 94.19 | Hurt |
| increase number of preceding routers r beyond six (ablation) | RDI-Net · CIFAR-10 (ResNet-110) | accuracy (%): 94.87 -> 94.41 | Hurt |
| increase SRM loss weight alpha (ablation) | RDI-Net · CIFAR-10 (ResNet-110) | GFLOPs | Improved |
Later papers that compared with this method
1 later accepted paper reports RDI-Net as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- SP-Net: Slowly Progressing Dynamic Inference Networks (ECCV 2022 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SP-Net: Slowly Progressing Dynamic Inference Networks (ECCV 2022 · similarity 0.89)
- A Graph Dynamics Prior for Relational Inference (AAAI 2024 · similarity 0.87)
- Memory-Augmented Dynamic Neural Relational Inference (ICCV 2021 · similarity 0.87)
- DS-Net++: Dynamic Weight Slicing for Efficient Inference in CNNs and Vision Transformers (TPAMI 2022 · similarity 0.86)
- HydraNets: Specialized Dynamic Architectures for Efficient Inference (CVPR 2018 · similarity 0.85)
- LightHGNN: Distilling Hypergraph Neural Networks into MLPs for 100x Faster Inference (ICLR 2024 · 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.
