Research map / Label-efficient and robust learning / Recognition and long-tail
This Looks Like Those: Illuminating Prototypical Concepts Using Multiple Visualizations
Chiyu Ma, Brandon Zhao, Chaofan Chen, Cynthia Rudin
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 |
|---|---|---|---|
| Replace single-patch prototypes with ball-shaped concepts shown through multiple training patches | ProtoPNet, ProtoPool, and TesNet · CUB 200-2011 test | accuracy: 77.9 ± 0.2 vs 78.0 ± 0.2; 78.2 ± 0.8 vs 77.9 ± 0.1; 81.5 ± 0.6 vs 80.3 ± 0.3 | No clear effect |
| Use multiple visualizations of a prototype | ProtoPNet explanations · User study class identification | accuracy: 51.5 ± 5.2 -> 62.1 ± 5.4 | Improved |
| Set radius initialization to 4.0 instead of 4.5 (ablation) | ProtoPool-Concepts · CUB 200-2011, ResNet-50 (iNat) | accuracy after pruning and fine-tuning: 85.2 ± 0.2 -> 81.5 ± 1.0 | Hurt |
| Set radius initialization to 5.0 instead of 4.5 (ablation) | ProtoPool-Concepts · CUB 200-2011, ResNet-50 (iNat) | accuracy after pruning and fine-tuning: 85.2 ± 0.2 -> 85.0 ± 0.3 | No clear effect |
| Use top-k cluster loss with k=1 rather than k=10 (ablation) | ProtoPool-Concepts · CUB 200-2011, ResNet-50 (iNat) | accuracy after fine-tuning: 85.2 ± 0.2 -> 67.7 ± 2.0 | Hurt |
| Increase top-k cluster loss k from 1 to 3 (ablation) | ProtoPool-Concepts · CUB 200-2011, ResNet-50 (iNat) | accuracy after fine-tuning: 67.7 ± 2.0 -> 83.5 ± 0.5 | Improved |
| Increase top-k cluster loss k from 3 to 5 (ablation) | ProtoPool-Concepts · CUB 200-2011, ResNet-50 (iNat) | accuracy after fine-tuning: 83.5 ± 0.5 -> 84.4 ± 0.3 | Improved |
| Increase top-k cluster loss k from 5 to 10 (ablation) | ProtoPool-Concepts · CUB 200-2011, ResNet-50 (iNat) | accuracy after fine-tuning: 84.4 ± 0.3 -> 85.2 ± 0.2 | Improved |
| Use top-k cluster loss (ablation) | ProtoPool-Concepts · CUB 200-2011, ResNet-50 (iNat) | accuracy before pruning: 85.2 ± 0.3; 85.3 ± 0.3; 85.5 ± 0.1; 85.3 ± 0.3 | No clear effect |
Closest papers by meaning
- Interpretable Image Classification with Adaptive Prototype-based Vision Transformers (NeurIPS 2024 · similarity 0.91)
- Interpretable Image Classification via Non-parametric Part Prototype Learning (CVPR 2025 · similarity 0.91)
- ProtoArgNet: Interpretable Image Classification with Super-Prototypes and Argumentation (AAAI 2025 · similarity 0.90)
- LucidPPN: Unambiguous Prototypical Parts Network for User-centric Interpretable Computer Vision (ICLR 2025 · similarity 0.90)
- Keep the Faith: Faithful Explanations in Convolutional Neural Networks for Case-Based Reasoning (AAAI 2024 · similarity 0.89)
- Improving Prototypical Visual Explanations with Reward Reweighing, Reselection, and Retraining (ICML 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.
