Research map / Adversarial robustness and security / Certified and empirical robustness
Robust Classification With Convolutional Prototype Learning
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 softmax classification with prototype matching trained by DCE | CNN image classifier · MNIST test | test accuracy (%): 99.08 -> 99.28 | Helped |
| Add prototype loss to CPL (ablation) | CPL (DCE) · MNIST test | test accuracy (%): 99.28 -> 99.33 | Helped |
| Vary prototype-loss weight λ among the reported settings (ablation) | GCPL (DCE+PL) · MNIST test | test accuracy (%): 99.33; 99.29; 99.30 | No clear effect |
| Replace softmax classification with prototype matching | CNN classifier · CIFAR-10 | accuracy: 90.26 -> 90.70; 91.37 -> 91.59; 91.32 -> 91.46; 92.50 -> 92.60 | Helped |
| Add prototype loss to CPL (ablation) | CPL · CIFAR-10 | accuracy: 90.70 -> 90.80; 91.59 -> 91.90; 91.46 -> 91.63; 92.60 -> 92.63 | Helped |
| Use MCE instead of the softmax loss in GCPL (ablation) | GCPL · OLHWDB | accuracy (%): 97.55 -> 97.35 | Hurt |
| Use MCL instead of the softmax loss in GCPL (ablation) | GCPL · OLHWDB | accuracy (%): 97.55 -> 97.61 | Helped |
| Use GMCL instead of the softmax loss in GCPL (ablation) | GCPL · OLHWDB | accuracy (%): 97.55 -> 97.36 | Hurt |
| Use DCE instead of the softmax loss in GCPL (ablation) | GCPL · OLHWDB | accuracy (%): 97.55 -> 97.58 | Helped |
| Use GCPL instead of softmax classification with fewer training samples | CNN classifier · MNIST test, reduced training data | test accuracy (%): 50%: 98.07±0.39 -> 99.12±0.10; 30%: 92.68±4.52 -> 98.89±0.10; 10%: 86.12±6.00 -> 97.80±0.22; 5%: 73.95±6.10 -> 96.44±0.4 | Helped |
| Increase the number of prototypes per class above 1 (ablation) | CPL (DCE), model C · CIFAR-10 | accuracy (%): 90.70 -> 90.40; 90.70 -> 90.37; 90.70 -> 90.67; 90.70 -> 90.46 | Hurt |
| Use distance-based confidence for rejection instead of softmax probability confidence | GCPL · MNIST acceptance and CIFAR-10 rejection | rejection rate (RR): At AR 99.20, GCPL Dist RR 99.99; softmax RR 98.02 | Helped |
Closest papers by meaning
- A Robust Prototype-Based Network with Interpretable RBF Classifier Foundations (AAAI 2025 · similarity 0.88)
- Variational Prototype Learning for Deep Face Recognition (CVPR 2021 · similarity 0.86)
- Hierarchically Robust Representation Learning (CVPR 2020 · similarity 0.86)
- Interpretable Image Classification with Adaptive Prototype-based Vision Transformers (NeurIPS 2024 · similarity 0.85)
- Adversarially Robust Few-shot Learning via Parameter Co-distillation of Similarity and Class Concept Learners (CVPR 2024 · similarity 0.84)
- Interpretable Image Classification via Non-parametric Part Prototype Learning (CVPR 2025 · 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.