Research map / Adversarial robustness and security / Attacks on text and graphs
Fooling SHAP with Stealthily Biased Sampling
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 output-space Wasserstein-regularized reweighting of existing background examples | SHAP sensitive-feature attribution · COMPAS and Marketing | relative decrease in sensitive-feature attribution amplitude: median reductions of about $90\% | Helped |
| Use output-space Wasserstein-regularized reweighting of existing background examples | SHAP sensitive-feature attribution · Adult Income and Communities | relative decrease in sensitive-feature attribution amplitude: median reduction in amplitude is about $50\% | Helped |
| Use output-space Wasserstein-regularized reweighting of existing background examples | SHAP sensitive-feature attribution · Adult Income and Communities | maximum relative decrease in sensitive-feature attribution amplitude: $90\% | Helped |
| Use Fool SHAP instead of brute-force or genetic background manipulation | SHAP sensitive-feature attribution · COMPAS, Adult Income, Marketing, Communities | reduction in sensitive-feature attribution | Helped |
| Select background samples using Fool SHAP while constraining detection rate | Audit detector · Four fairness datasets | fraud detection | No clear effect |
| Calibrate the detector using uniformly sampled subsets | Audit detector · COMPAS, Adult, Marketing, Communities | false positive rate: 4.0, 4.6, 4.0; 4.3, 4.3, 4.2; 4.9, 5.0; 3.8, 4.2 | No clear effect |
Closest papers by meaning
- Shaping Up SHAP: Enhancing Stability through Layer-Wise Neighbor Selection (AAAI 2024 · similarity 0.85)
- Auditing Local Explanations is Hard (NeurIPS 2024 · similarity 0.84)
- Provably Accurate Shapley Value Estimation via Leverage Score Sampling (ICLR 2025 · similarity 0.83)
- X-Hacking: The Threat of Misguided AutoML (ICML 2025 · similarity 0.83)
- Nearly Tight Black-Box Auditing of Differentially Private Machine Learning (NeurIPS 2024 · similarity 0.83)
- Auditing Privacy Mechanisms via Label Inference Attacks (NeurIPS 2024 · similarity 0.82)
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.