Research map / Label-efficient and robust learning / Recognition and long-tail
IG 2 : Integrated Gradient on Iterative Gradient Path for Feature Attribution
Yue Zhuo, Zhiqiang Ge
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 GradCF baseline instead of black baseline on straight-line path (ablation) | Vanilla IG · WM-811K wafer map | ground truth SUM: 0.342 -> 0.510 | Improved |
| add GradCF baseline instead of train data baseline on straight-line path (ablation) | Vanilla IG · WM-811K wafer map | SIC AUC-ADD: 0.777 -> 0.842 | Improved |
| add GradCF baseline instead of black baseline on Guided IG path (ablation) | Guided IG · WM-811K wafer map | SIC AUC-DEL: 0.050 -> 0.037 | Improved |
| add GradCF baseline instead of train data baseline on Guided IG path (ablation) | Guided IG · WM-811K wafer map | ground truth AUC: 0.804 -> 0.781 | Mixed |
| use GradPath instead of straight-line path with GradCF baseline (ablation) | IG2 · WM-811K wafer map | SIC AUC-ADD: 0.842 -> 0.898 | Improved |
| use GradPath instead of Guided IG path with GradCF baseline (ablation) | IG2 · WM-811K wafer map | ground truth SUM: 0.472 -> 0.551 | Improved |
| use relevant instead of irrelevant references (ablation) | IG2 · ImageNet (images with interference objects) | SIC AUC-ADD: 0.876 -> 0.901 | Improved |
| use relevant instead of irrelevant references (ablation) | IG2 · ImageNet (pure background images) | SIC AUC-ADD: 0.568 -> 0.537 | Hurt |
| use relevant instead of irrelevant references (ablation) | IG2 · ImageNet (whole set) | ground truth AUC: 0.805 -> 0.798 | No clear effect |
| use IG2 instead of Expected IG | Expected IG · ImageNet | SIC AUC-ADD: 0.557 -> 0.656 | Improved |
| use IG2 instead of Expected IG | Expected IG · ImageNet | SIC AUC-DEL: 0.116 -> 0.115 | No clear effect |
| use IG2 instead of Vanilla IG | Vanilla IG · XAI-BENCH synthetic | faithfulness: 0.349 -> 0.610 | Improved |
Closest papers by meaning
- Integrated Decision Gradients: Compute Your Attributions Where the Model Makes Its Decision (AAAI 2024 · similarity 0.93)
- Manifold Integrated Gradients: Riemannian Geometry for Feature Attribution (ICML 2024 · similarity 0.87)
- Iterative Search Attribution for Deep Neural Networks (ICML 2024 · similarity 0.85)
- Denoising Diffusion Path: Attribution Noise Reduction with An Auxiliary Diffusion Model (NeurIPS 2024 · similarity 0.83)
- Attack-free Evaluating and Enhancing Adversarial Robustness on Categorical Data (ICML 2024 · similarity 0.83)
- Derivative-Free Diffusion Manifold-Constrained Gradient for Unified XAI (CVPR 2025 · 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.
