Research map / Architectures and efficiency / RNNs and normalization
Why Capsule Neural Networks Do Not Scale: Challenging the Dynamic Parse-Tree Assumption
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 |
|---|---|---|---|
| replace RBA with uniform routing (ablation) | CapsNet with 64 capsules, dim 8, depth 2 · AffNIST validation | accuracy: 0.88 -> 0.92 | Helped |
| replace RBA with uniform routing (ablation) | CapsNet with 32 capsules, dim 64, depth 2 · CIFAR10 validation | accuracy: 0.79 -> 0.79 | No clear effect |
| increase capsule depth from 1 to 6, selecting the best model at each depth (ablation) | CapsNet on AffNIST · AffNIST validation | accuracy: 0.88 -> 0.83 | Hurt |
| increase capsule depth from 2 to 4, selecting the best model at each depth (ablation) | CapsNet on CIFAR10 · CIFAR10 validation | accuracy: 0.79 -> 0.69 | Hurt |
| apply stronger affine transformations to the input | CapsNet capsule activations · AffNIST | activation correlation | Hurt |
| use RBA rather than uniform routing (ablation) | CapsNet on AffNIST · AffNIST validation | accuracy: 0.92 -> 0.88 | Hurt |
| add intermediate capsule layers with RBA routing (ablation) | CapsNet on AffNIST · AffNIST validation | dead capsule rate: 0.00 -> 0.63 ± 0.04 | Hurt |
| add intermediate capsule layers with RBA routing (ablation) | CapsNet on CIFAR10 · CIFAR10 validation | dead capsule rate: 0.00 -> 0.81 | Hurt |
Closest papers by meaning
- STAR-Caps: Capsule Networks with Straight-Through Attentive Routing (NeurIPS 2019 · similarity 0.88)
- A Receptor Skeleton for Capsule Neural Networks (ICML 2021 · similarity 0.88)
- ParseCaps: An Interpretable Parsing Capsule Network for Medical Image Diagnosis (AAAI 2025 · similarity 0.88)
- Ladder Capsule Network (ICML 2019 · similarity 0.87)
- DeepCaps: Going Deeper With Capsule Networks (CVPR 2019 · similarity 0.87)
- OrthCaps: An Orthogonal CapsNet with Sparse Attention Routing and Pruning (CVPR 2024 · similarity 0.87)
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.