Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Adversarial Attacks on Linear Contextual Bandits
Evrard Garcelon, Baptiste Roziere, Laurent Meunier, Jean Tarbouriech, Olivier Teytaud, Alessandro Lazaric, Matteo Pirotta
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 |
|---|---|---|---|
| Apply ContextualConic context attack to LinUCB | LinUCB · Synthetic, Jester, and MovieLens | target-set selection rate: 28.91% -> 98.55%; 26.59% -> 98.36%; 31.13% -> 99.61% | Improved |
| Apply ContextualConic context attack to ε-greedy | ε-greedy · Synthetic, Jester, and MovieLens | target-set selection rate: 25.7% -> 89.71%; 25.85% -> 99.85%; 31.78% -> 99.92% | Improved |
| Apply ContextualConic context attack to LinTS | LinTS · Synthetic | target-set selection rate: 27.2% -> 30.93% | No clear effect |
| Apply ContextualConic context attack to LinTS | LinTS · Jester and MovieLens | target-set selection rate: 26.10% -> 97.26%; 33.24% -> 98.82% | Improved |
| Use full rather than relaxed optimization for the offline LinUCB context attack | LinUCB offline context attack · Synthetic and MovieLens | target-set selection rate: 13.76% -> 88.30%; 4.09% -> 99.99% | Improved |
| Use full rather than relaxed optimization for the offline LinUCB context attack | LinUCB offline context attack · Jester | target-set selection rate: 97.81% -> 99.98% | Improved |
| Apply full offline context attack | ε-greedy · Synthetic, Jester, and MovieLens | target-set selection rate: 0.01% -> 99.98%; 0.00% -> 99.95%; 0.03% -> 99.97% | Improved |
| Apply relaxed offline context attack | LinTS · Synthetic, Jester, and MovieLens | target-set selection rate: 0.02% -> 18.21%; 0.01% -> 80.48%; 0.05% -> 5.56% | Mixed |
| Apply CACE reward attack | LinUCB · Synthetic, Jester, and Movilens | target arm draws: 86, 731.6 -> 996, 238.6; 23, 548.16 -> 921, 083.69; 25, 017.31 -> 944, 721.28 | Improved |
| Use stationary CACE rather than CACE reward attack | LinUCB · Synthetic, Jester, and Movilens | target arm draws: 996, 238.6 -> 995, 578.88; 921, 083.69 -> 862, 095.67; 944, 721.28 -> 931, 531.6 | Mixed |
| Attack rewards with CACE | contextual bandit algorithms · Synthetic data | target-arm selection rate: more than 99.4%; more than 97.8% | Improved |
| Attack rewards with CACE | contextual bandit algorithms · Jester and MovieLens | target-arm selection rate: more than 94.0%; more than 77.0% | Improved |
Closest papers by meaning
- Adversarial Attacks on Stochastic Bandits (NeurIPS 2018 · similarity 0.89)
- Robust Neural Contextual Bandit against Adversarial Corruptions (NeurIPS 2024 · similarity 0.89)
- An Improved Algorithm for Adversarial Linear Contextual Bandits via Reduction (NeurIPS 2025 · similarity 0.88)
- Nearly Optimal Algorithms for Contextual Dueling Bandits from Adversarial Feedback (ICML 2025 · similarity 0.88)
- Practical Adversarial Attacks on Stochastic Bandits via Fake Data Injection (NeurIPS 2026 · similarity 0.87)
- Adversaries With Incentives: A Strategic Alternative to Adversarial Robustness (ICLR 2025 · 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.
