Research map / Label-efficient and robust learning / Recognition and long-tail
iCaps: An Interpretable Classifier via Disentangled Capsule Networks
Dahuin Jung, Jonghyun Lee, Jihun Yi, Sungroh Yoon
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 |
|---|---|---|---|
| use iCaps instead of ResNet-18 classifier | iCaps · MNIST | accuracy of p(y|c): 0.992 -> 0.992 | No clear effect |
| use iCaps instead of DeepCaps | iCaps · SVHN | accuracy of p(y|c): 0.971 -> 0.920 | Hurt |
| use iCaps instead of ResNet-18 | iCaps · CelebA | accuracy of p(y|c): 0.977 -> 0.984 | Improved |
| use class-relevant intra-class variation in c (iCaps) instead of LORD | disentanglement · SVHN | accuracy of p(y|r): 0.163 -> 0.099 | Improved |
| use iCaps instead of ML-VAE | disentanglement · MNIST | accuracy of p(y|r): 0.717 -> 0.099 | Improved |
| use iCaps instead of Cycle-VAE | disentanglement · CelebA | accuracy of p(y|r): 0.793 -> 0.501 | Improved |
| add L_concept loss (ablation) | iCaps · MNIST/SVHN/CelebA | mutual information I(c;y) | Improved |
| add concept-distinctness discriminator D_CR (ablation) | iCaps · MNIST | distinctness of class capsule elements | Improved |
Closest papers by meaning
- Interpretable Graph Capsule Networks for Object Recognition (AAAI 2021 · similarity 0.88)
- STAR-Caps: Capsule Networks with Straight-Through Attentive Routing (NeurIPS 2019 · similarity 0.87)
- Conditional Variational Capsule Network for Open Set Recognition (ICCV 2021 · similarity 0.86)
- ParseCaps: An Interpretable Parsing Capsule Network for Medical Image Diagnosis (AAAI 2025 · similarity 0.86)
- NeSyFOLD: A Framework for Interpretable Image Classification (AAAI 2024 · similarity 0.84)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 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.
