Research map / Architectures and efficiency / RNNs and normalization
Relative Attributing Propagation: Interpreting the Comparative Contributions of Individual Units in Deep Neural Networks
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 |
|---|---|---|---|
| Use RAP instead of LRP-α₂β₁ | VGG-16 explanations on ImageNet · ImageNet, Outside-Inside Ratio (ALL) | Outside-Inside Ratio: 0.616 -> 0.252 | Helped |
| Use RAP instead of LRP-α₁β₀ | VGG-16 explanations on ImageNet · ImageNet, Outside-Inside Ratio (POS) | Outside-Inside Ratio: 0.474 -> 0.341 | Helped |
| Use RAP instead of LRP-α₂β₁ | ResNet-50 explanations on ImageNet · ImageNet, Outside-Inside Ratio (ALL) | Outside-Inside Ratio: 0.302 -> 0.164 | Helped |
| Use RAP instead of LRP-α₂β₁ | ResNet-50 explanations on ImageNet · ImageNet, Outside-Inside Ratio (POS) | Outside-Inside Ratio: 0.299 -> 0.166 | Helped |
| Use RAP instead of LRP-α₁β₀ | VGG-16 explanations · ImageNet segmentation | mIOU: 55.78 -> 62.23 | Helped |
| Use RAP instead of LRP-α₁β₀ | VGG-16 explanations · Pascal VOC 2012 segmentation | mIOU: 49.82 -> 55.60 | Helped |
| Use RAP instead of Xiong et al.'s method | ImageNet generic object segmentation · ImageNet segmentation | mIOU: 64.22 -> 62.23 | Hurt |
| Assign relevance according to relative influence rather than contribution sign | RAP explanations · ImageNet segmentation | mIOU: 55.78 -> 62.23 | Helped |
| Use positive and negative relevance to distinguish relevant from irrelevant pixels | RAP explanations · ImageNet and Pascal VOC 2012 | mIOU: ImageNet: 55.78 -> 62.23; Pascal VOC: 49.82 -> 55.60 | Helped |
| Assign negative relevance to low-influence regions using RAP | VGG-16 and ResNet-50 explanations · ImageNet validation images, least-relevant-first perturbation | Prediction accuracy under negative-attribution removal | Helped |
Closest papers by meaning
- Interpreting Deep Neural Networks with Relative Sectional Propagation by Analyzing Comparative Gradients and Hostile Activations (AAAI 2021 · similarity 0.92)
- Bias Also Matters: Bias Attribution for Deep Neural Network Explanation (ICML 2019 · similarity 0.87)
- How to Probe: Simple Yet Effective Techniques for Improving Post-hoc Explanations (ICLR 2025 · similarity 0.85)
- Evaluating Neuron Explanations: A Unified Framework with Sanity Checks (ICML 2025 · similarity 0.84)
- What Variables Affect Out-of-Distribution Generalization in Pretrained Models? (NeurIPS 2024 · similarity 0.84)
- MFABA: A More Faithful and Accelerated Boundary-Based Attribution Method for Deep Neural Networks (AAAI 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.