Research map / Architectures and efficiency / CNN design
Training Interpretable Convolutional Neural Networks by Differentiating Class-specific Filters
Haoyu Liang, Zhihao Ouyang, Yuyuan Zeng, Hang Su, Zihao He, Shu-Tao Xia, Jun Zhu, Bo Zhang
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 |
|---|---|---|---|
| add class-specific gate training | ResNet20 · CIFAR-10 | accuracy: 0.9169 -> 0.9192 | No clear effect |
| add class-specific gate training | ResNet18 · ImageNet | accuracy: 0.6976 -> 0.6784 | Hurt |
| add class-specific gate training | ResNet152 · PASCAL VOC 2010 | accuracy: 0.8447 -> 0.8518 | No clear effect |
| add class-specific gate training | ResNet20 · CIFAR-10 | MIS: 0.1119 -> 0.1603 | Improved |
| add class-specific gate training | ResNet18 · ImageNet | MIS: 0.5514 -> 0.6259 | Improved |
| add class-specific gate training | ResNet152 GradMap · PASCAL VOC | Avg-IoU: 0.2606 -> 0.3035 | Improved |
| add class-specific gate training | ResNet152 ActivMap · PASCAL VOC | AP30: 0.6759 -> 0.8251 | Improved |
| add class-specific gate training | ResNet152 CAM · PASCAL VOC | Avg-IoU: 0.3483 -> 0.3694 | Improved |
| add class-specific gate training | ResNet20 FGSM detection · CIFAR-10 (500 samples) | error rate (%): 4.76 -> 4.50 | Improved |
| add class-specific gate training | ResNet20 PGD detection · CIFAR-10 (2000 samples) | error rate (%): 4.08 -> 3.25 | Improved |
| add class-specific gate training | ResNet20 CW detection · CIFAR-10 (500 samples) | error rate (%): 7.33 -> 7.03 | Improved |
| add class-specific gate training | ResNet20 single pixel attack · CIFAR-10 | attack success rate (%): 14.00% -> 2.00% | Improved |
Closest papers by meaning
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.90)
- Interpretable CNNs for Object Classification (TPAMI 2020 · similarity 0.89)
- Neural Representations Reveal Distinct Modes of Class Fitting in Residual Convolutional Networks (AAAI 2023 · similarity 0.87)
- Interpretable Convolutional Neural Networks (CVPR 2018 · similarity 0.87)
- B-cosification: Transforming Deep Neural Networks to be Inherently Interpretable (NeurIPS 2024 · similarity 0.85)
- From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit (NeurIPS 2025 · similarity 0.85)
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.
