Research map / Label-efficient and robust learning / Self-supervised and contrastive
Visualization of Supervised and Self-Supervised Neural Networks via Attribution Guided Factorization
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 |
|---|---|---|---|
| combine gradient propagation, attribution propagation, and guided factorization | AGF · ImageNet negative perturbation, predicted class | AUC: 38.7 -> 38.9 | Helped |
| combine gradient propagation, attribution propagation, and guided factorization | AGF · ImageNet negative perturbation, target class | AUC: 39.4 -> 40.0 | Helped |
| combine gradient propagation, attribution propagation, and guided factorization | AGF · ImageNet-Segmentation | Pixel Acc.: 78.8 -> 79.9 | Helped |
| combine gradient propagation, attribution propagation, and guided factorization | AGF · ImageNet-Segmentation | mAP: 73.7 -> 76.7 | Helped |
| combine gradient propagation, attribution propagation, and guided factorization | AGF · VOC'12 | Pixel Acc.: 72.8 -> 77.5 | Helped |
| combine gradient propagation, attribution propagation, and guided factorization | AGF · VOC'12 | mAP: 41.7 -> 45.9 | Helped |
| add the input-agnostic attribution A (ablation) | AGF · ImageNet negative perturbation | AUC: 34.6 -> 38.9 | Helped |
| add the feature-map factorization F_x (ablation) | AGF · ImageNet negative perturbation | AUC: 37.5 -> 38.9 | Helped |
| add the gradient factorization F_∇x (ablation) | AGF · ImageNet negative perturbation | AUC: 37.1 -> 38.9 | Helped |
| add the input-gradient interaction M (ablation) | AGF · ImageNet negative perturbation | AUC: 37.2 -> 38.9 | Helped |
| add attribution-dependent weighting of the feature-map and interaction residual (ablation) | AGF · ImageNet negative perturbation | AUC: 37.5 -> 38.9 | Helped |
| replace the guided attribution residual with a Grad-CAM residual (ablation) | AGF · ImageNet negative perturbation | AUC: 38.9 -> 37.1 | Hurt |
Closest papers by meaning
- Rethinking Interpretation: Input-Agnostic Saliency Mapping of Deep Visual Classifiers (AAAI 2023 · similarity 0.88)
- Explaining Convolutional Neural Networks through Attribution-Based Input Sampling and Block-Wise Feature Aggregation (AAAI 2021 · similarity 0.87)
- Grid Saliency for Context Explanations of Semantic Segmentation (NeurIPS 2019 · similarity 0.87)
- How to Probe: Simple Yet Effective Techniques for Improving Post-hoc Explanations (ICLR 2025 · similarity 0.87)
- There and Back Again: Revisiting Backpropagation Saliency Methods (CVPR 2020 · similarity 0.87)
- Don't trust your eyes: on the (un)reliability of feature visualizations (ICML 2024 · similarity 0.86)
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.