Research map / Federated learning and privacy / Privacy attacks and secure learning
XRand: Differentially Private Defense against Explanation-Guided Attacks
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 |
|---|---|---|---|
| Apply XRand to SHAP explanations | LightGBM · EMBER; XBA with 1% poison rate; ε = 1.0 | attack success rate: 77.8% -> 10.2% | Helped |
| Apply XRand to SHAP explanations | EmberNN · EMBER; XBA with 1% poison rate; ε = 10.0 | attack success rate: 5.3% | Helped |
| Apply XRand to SHAP explanations | Random Forest · Contagio PDF; XBA with 1% poison rate; ε = 10.0 | attack success rate: 9.6% | Helped |
| Apply XRand to SHAP explanations | Linear SVM · Drebin; XBA with 1% poison rate; ε = 10.0 | attack success rate: 7% | Helped |
| Tighten the privacy budget ε | XRand · EMBER, Contagio PDF, and Drebin; XBA | attack success rate | Helped |
| Increase the portion of poisoned samples | XBA against LightGBM and EmberNN · EMBER | attack success rate | Mixed |
| Use XRand + PixelDP rather than XRand + boosting randomized smoothing | XRand inference-time robustness · EMBER; ε ∈ [0.1, 1.0] | certified accuracy: 89.17% and 90.42% compared with 50% | Helped |
| Reduce ε | XRand training-time certified robustness · EMBER; ε ∈ [0.1, 100.0] | certified accuracy | Helped |
| Increase the poisoning-training threshold r_tr | XRand training-time certified robustness · EMBER | certified accuracy | Hurt |
| Use XRand explanations with ε = 1.0 or 10.0 instead of original SHAP explanations | SHAP explanations · EMBER; log-odds faithfulness | log-odds | No clear effect |
Closest papers by meaning
- X-Hacking: The Threat of Misguided AutoML (ICML 2025 · similarity 0.84)
- ExpProof : Operationalizing Explanations for Confidential Models with ZKPs (ICML 2025 · similarity 0.84)
- Auditing Private Prediction (ICML 2024 · similarity 0.82)
- Provable Robustness against Backdoor Attacks via the Primal-Dual Perspective on Differential Privacy (NeurIPS 2026 · similarity 0.82)
- LT-Defense: Searching-free Backdoor Defense via Exploiting the Long-tailed Effect (NeurIPS 2024 · similarity 0.81)
- From Counterfactuals to Trees: Competitive Analysis of Model Extraction Attacks (NeurIPS 2025 · similarity 0.81)
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.