Research map / Architectures and efficiency / RNNs and normalization
Learning Propagation Rules for Attribution Map Generation
Yiding Yang, Jiayan Qiu, Mingli Song, Dacheng Tao, Xinchao Wang
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 |
|---|---|---|---|
| use learned propagation rules instead of random pixel removal order | attribution map · ROAR | test accuracy (%) after removal: 73.65% -> 71.83% at 40%; 59.58% -> 50.72% at 90% | Improved |
| use learned propagation rules instead of SQ_SG | attribution map · ROAR | test accuracy (%) after removal: 74.90% -> 71.83% at 40%; 57.47% -> 50.72% at 90% | Improved |
| add sign-aware loss (ablation) | learnable attribution plugin · ImageNet | LeRF curve | Improved |
| add sign-aware loss (ablation) | learnable attribution plugin · ImageNet | MoRF curve | Hurt |
| add reference baseline (ablation) | learnable attribution plugin · ImageNet | MoRF and LeRF | No clear effect |
| use standard convolution as plugin module | learnable attribution plugin · ImageNet | attribution map quality | Hurt |
| increase lambda loss balance weight | Ours · ImageNet VGG-16 | MoRF AUC: 0.23 -> 0.21 | No clear effect |
| increase gamma from 1 to 21 | Ours · ImageNet VGG-16 | MoRF AUC: 0.40 -> 0.13 | Improved |
| increase gamma from 21 to 60 | Ours · ImageNet VGG-16 | MoRF AUC: 0.13 -> 0.22 | Hurt |
| run more optimization iterations (1 to 54) | Ours · ImageNet VGG-16 | MoRF AUC: 0.86 -> 0.22 | Hurt |
| raise learning rate from 0.01 to 0.45 | Ours · ImageNet VGG-16 | MoRF AUC: 0.40 -> 0.20 | Improved |
| raise alpha from 0.55 to 0.86 | Ours · ImageNet VGG-16 | MoRF AUC: 0.47 -> 0.18 | Improved |
Closest papers by meaning
- Generating Attribution Maps With Disentangled Masked Backpropagation (ICCV 2021 · similarity 0.89)
- Pixel-level Certified Explanations via Randomized Smoothing (ICML 2025 · similarity 0.86)
- Benchmarking the Attribution Quality of Vision Models (NeurIPS 2024 · similarity 0.85)
- Data Attribution for Text-to-Image Models by Unlearning Synthesized Images (NeurIPS 2024 · similarity 0.84)
- Diffusion Models for Attribution (AAAI 2025 · similarity 0.83)
- Elucidating the design space of classifier-guided diffusion generation (ICLR 2024 · similarity 0.83)
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.
