Research map / Architectures and efficiency / RNNs and normalization
Ladder Capsule Network
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 dynamic routing with pruning layer + ladder layer (L-CapsNet) | CapsNet · affNIST | test error (%): 21.0 -> 12.5 (K=70, 9x9 kernel) | Helped |
| replace dynamic routing with pruning layer + ladder layer (L-CapsNet) | CapsNet · MNIST | test error (%): 0.25 -> 0.50 (K=70, 9x9 kernel) | Hurt |
| pruning to K=50 instead of K=70 most active capsules (ablation) | L-CapsNet · MNIST | test error (%): 0.50 -> 0.74 | Hurt |
| pruning to K=100 instead of K=70 most active capsules (ablation) | L-CapsNet · MNIST | test error (%): 0.50 -> 0.80 | Hurt |
| pruning to K=100 instead of K=70 most active capsules (ablation) | L-CapsNet · affNIST | test error (%): 12.5 -> 13.2 | Hurt |
| pruning to K=50 instead of K=70 most active capsules (ablation) | L-CapsNet · affNIST | test error (%): 12.5 -> 13.0 | Hurt |
| use 15x15 kernels instead of 9x9 in the L-CapsNet (ablation) | L-CapsNet · affNIST | test error (%): K=70: 12.5 -> 12.2; K=50: 13.0 -> 12.5; K=100: 13.2 -> 13.1 | Mixed |
| use 15x15 kernels instead of 9x9 in the L-CapsNet (ablation) | L-CapsNet · MNIST | test error (%): K=70: 0.50 -> 0.73 (worse); K=50: 0.74 -> 0.69 (better) | Mixed |
| pruning layer (selecting K most active capsules) vs. full routing; reduces training time per iteration | CapsNet · MNIST training time per iteration | seconds: 1.732 (CapsNet r=3) -> 0.2034 (L-CapsNet K=50) | Helped |
Closest papers by meaning
- Introducing Routing Uncertainty in Capsule Networks (NeurIPS 2020 · similarity 0.90)
- Building Deep Equivariant Capsule Networks (ICLR 2020 · similarity 0.90)
- Self-Routing Capsule Networks (NeurIPS 2019 · similarity 0.90)
- Dynamic Routing Between Capsules (NeurIPS 2017 · similarity 0.89)
- Matrix capsules with EM routing (ICLR 2018 · similarity 0.89)
- OrthCaps: An Orthogonal CapsNet with Sparse Attention Routing and Pruning (CVPR 2024 · similarity 0.89)
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.