Research map / Architectures and efficiency / CNN design
Interpretable Neural Network Decoupling
Yuchao Li, Rongrong Ji, Shaohui Lin, Baochang Zhang, Chenqian Yan, Yongjian Wu, Feiyue Huang, Ling Shao
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 mutual-information loss L_mi (ablation) | ResNet-56 decoupling · CIFAR-10 | FLOPs: 117M -> 63M | Improved |
| add mutual-information loss L_mi (ablation) | ResNet-56 decoupling · CIFAR-10 | dynamic filters: 0.00% -> 30.27% | Improved |
| add mutual-information loss L_mi (ablation) | ResNet-56 decoupling · CIFAR-10 | top-1 accuracy: 92.81 -> 93.08 | Improved |
| add KL loss L_kl (ablation) | ResNet-56 decoupling · CIFAR-10 | top-1 accuracy: 92.99 -> 93.08 | Improved |
| add KL loss L_kl (ablation) | ResNet-56 decoupling · CIFAR-10 | energetic filters: 38.45% -> 34.49% | Improved |
| add sparsity loss L_s to ACM (ablation) | ResNet-56 decoupling · CIFAR-10 | FLOPs: 118M -> 63M | Improved |
| add sparsity loss L_s to ACM (ablation) | ResNet-56 decoupling · CIFAR-10 | top-1 accuracy: 93.17 -> 92.94 | Hurt |
| decouple VGG-16 architecture (ablation) | VGG-16 · BRODEN Conv5_3 | interpretability score: 0.0787 -> 0.0879 | Improved |
| decouple ResNet-18 architecture | ResNet-18 · ImageNet 2012 val | top-1 accuracy: 69.76 -> 67.62 | Hurt |
| decouple ResNet-18 architecture | ResNet-18 · BRODEN Block1 | interpretability score: 0.0527 -> 0.1062 | Improved |
| use path-level multiple filters instead of single filter for semantic features | decoupled ResNet-18 · BRODEN | representation ability: about 3x | Improved |
| use decoupling instead of FBS for pruning | VGG-16 · ImageNet 2012 val | top-5 accuracy: 89.86 -> 90.32 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ThiNet — ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression (ICCV 2017) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.88)
- Decoupled Networks (CVPR 2018 · similarity 0.87)
- Network Dissection: Quantifying Interpretability of Deep Visual Representations (CVPR 2017 · similarity 0.87)
- Interpreting Deep Visual Representations via Network Dissection (TPAMI 2018 · similarity 0.87)
- Deeply Explain CNN Via Hierarchical Decomposition (IJCV 2023 · similarity 0.86)
- Network Inversion of Convolutional Neural Nets (Student Abstract) (AAAI 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.
