Research map / Architectures and efficiency / RNNs and normalization
Attribution Preservation in Network Compression for Reliable Network Interpretation
Geondo Park, June Yong Yang, Sung Ju Hwang, Eunho Yang
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 stochastic sensitivity-weighted attribution matching during VGG16/8 distillation | VGG16/8 distilled from VGG16 · PASCAL VOC 2012 validation | AUC: 78.74 -> 88.06 | Improved |
| Add stochastic sensitivity-weighted attribution matching during VGG16/8 distillation | VGG16/8 distilled from VGG16 · PASCAL VOC 2012 validation | mAP: 76.91 -> 80.86 | Improved |
| Replace naive fine-tuning with stochastic sensitivity-weighted matching after unstructured pruning | VGG16 · PASCAL VOC 2012 validation | AUC: 84.13 -> 88.45 | Improved |
| Replace naive fine-tuning with stochastic sensitivity-weighted matching after structured pruning | VGG16 · PASCAL VOC 2012 validation | AUC: 79.98 -> 88.34 | Improved |
| Add stochastic sensitivity-weighted matching after 60% unstructured pruning | VGG16 · ImageNet segmentation subset | AUC: 76.01 -> 80.88 | Improved |
| Add stochastic sensitivity-weighted matching after structured pruning | ResNet34 · ImageNet segmentation subset | AUC: 81.20 -> 84.65 | Improved |
| Use sensitivity-weighted matching rather than equally weighted activation matching (ablation) | VGG16/8 distillation · PASCAL VOC 2012 validation | AUC: 81.99 -> 87.88 | Improved |
| Add stochastic channel masking to sensitivity-weighted matching (ablation) | VGG16/8 distillation · PASCAL VOC 2012 validation | AUC: 87.88 -> 88.06 | Improved |
| Use sensitivity-weighted matching rather than equally weighted matching (ablation) | VGG16/8 distillation · PASCAL VOC 2012 validation | Cosine similarity: 0.652 -> 0.813 | Improved |
| Use stochastic sensitivity-weighted Grad-CAM matching | VGG16/8 distillation · PASCAL VOC 2012 validation | AUC for Excitation Backprop: 76.31 -> 82.31 | Improved |
| Use stochastic sensitivity-weighted Grad-CAM matching | VGG16/8 distillation · PASCAL VOC 2012 validation | AUC for LRP: 79.60 -> 82.46 | Improved |
| Use stochastic sensitivity-weighted Grad-CAM matching | VGG16/8 distillation · PASCAL VOC 2012 validation | Point accuracy for RAP: 56.87 -> 65.66 | Improved |
Closest papers by meaning
- Evaluating Attribution for Graph Neural Networks (NeurIPS 2020 · similarity 0.86)
- Axiomatic Attribution for Deep Networks (ICML 2017 · similarity 0.85)
- Towards better understanding of gradient-based attribution methods for Deep Neural Networks (ICLR 2018 · similarity 0.85)
- Exploiting the Relationship Between Kendall's Rank Correlation and Cosine Similarity for Attribution Protection (NeurIPS 2022 · similarity 0.85)
- A Practical Upper Bound for the Worst-Case Attribution Deviations (CVPR 2023 · similarity 0.85)
- Enhanced Regularizers for Attributional Robustness (AAAI 2021 · 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.
