Research map / Adversarial robustness and security / Certified and empirical robustness
Evaluations and Methods for Explanation through Robustness Analysis
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 |
|---|---|---|---|
| optimize feature sets using subset-restricted adversarial robustness and set-aggregated greedy selection | Greedy-AS explanations · MNIST | Robustness-\overline{S_r} AUC: 93.24 -> 98.01 | Helped |
| optimize feature sets using subset-restricted adversarial robustness and set-aggregated greedy selection | Greedy-AS explanations · MNIST | Robustness-S_r AUC: 91.72 -> 82.81 | Helped |
| optimize feature sets using subset-restricted adversarial robustness and set-aggregated greedy selection | Greedy-AS explanations · ImageNet | Robustness-\overline{S_r} AUC: 27.13 -> 31.62 | Helped |
| optimize feature sets using subset-restricted adversarial robustness and set-aggregated greedy selection | Greedy-AS explanations · ImageNet | Robustness-S_r AUC: 45.53 -> 43.97 | Helped |
| optimize feature sets using subset-restricted adversarial robustness and set-aggregated greedy selection | Greedy-AS explanations · Yahoo!Answer | Robustness-\overline{S_r} AUC: 1.97 -> 2.13 | Helped |
| optimize feature sets using subset-restricted adversarial robustness and set-aggregated greedy selection | Greedy-AS explanations · Yahoo!Answer | Robustness-S_r AUC: 2.91 -> 2.41 | Helped |
| replace set-aggregated scores with pure greedy marginal feature selection (ablation) | Greedy-AS · MNIST | Robustness-\overline{S_r} AUC: 98.01 -> 83.57 | Hurt |
| replace set-aggregated scores with pure greedy marginal feature selection (ablation) | Greedy-AS · ImageNet | Robustness-S_r AUC: 43.97 -> 58.45 | Hurt |
| use one Banzhaf regression step rather than iterative selection (ablation) | Greedy-AS · MNIST | Robustness-\overline{S_r} AUC: 98.01 -> 86.37 | Hurt |
| use one Banzhaf regression step rather than iterative selection (ablation) | Greedy-AS · ImageNet | Robustness-S_r AUC: 43.97 -> 47.07 | Hurt |
| evaluate zero-baseline IG using a zero insertion reference instead of a random reference | IG on MNIST · MNIST | Insertion AUC: 177.12 -> 802.57 | Helped |
| evaluate zero-baseline SHAP using a reference of 0.75 instead of a random reference | SHAP on MNIST · MNIST | Insertion AUC: 125.93 -> 79.28 | Hurt |
Closest papers by meaning
- Robust Explanation for Free or At the Cost of Faithfulness (ICML 2023 · similarity 0.91)
- Promoting Counterfactual Robustness through Diversity (AAAI 2024 · similarity 0.88)
- Interpretation of Neural Networks Is Fragile (AAAI 2019 · similarity 0.88)
- Indicators of Attack Failure: Debugging and Improving Optimization of Adversarial Examples (NeurIPS 2022 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- Precise Accuracy / Robustness Tradeoffs in Regression: Case of General Norms (ICML 2024 · similarity 0.87)
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.