Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Coordinated Attacks against Contextual Bandits: Fundamental Limits and Defense Mechanisms
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 robust estimation across users instead of ignoring adversarial users | Multitask contextual bandits · Synthetic contextual bandits | policy sub-optimality | Helped |
| Increase the number of contexts or actions | Robust MCB · Synthetic contextual bandits | policy sub-optimality | No clear effect |
| Use user identifiers to distinguish data sources rather than applying corruption-robust bandits without that information | Contextual bandits with coordinated adversaries · Synthetic contextual bandits | policy robustness | Helped |
| Increase the number of users up to the robust-estimation threshold | Robust MCB · Synthetic contextual bandits | policy quality | Helped |
| Increase the number of users beyond the threshold | Robust MCB · Synthetic contextual bandits | policy quality | No clear effect |
| Overestimate the adversarial fraction by setting the algorithm's α above the actual rate (ablation) | Robust MCB · Synthetic contextual bandits with actual corruption rate below 0.1 | policy robustness | No clear effect |
| Underestimate the adversarial fraction by setting the algorithm's α below the actual rate (ablation) | Robust MCB · Synthetic contextual bandits with actual corruption rate above 0.1 | policy robustness | Hurt |
| Add small random perturbations to good users' mean rewards (ablation) | Robust MCB · Synthetic contextual bandits with similar but nonidentical preferences | policy quality | Mixed |
| Replace shared-preference users with arbitrarily coordinated adversarial users | Multitask contextual bandits · Uniform-context lower-bound construction | per-user interactions required: \tilde{\Omega}(\min(S,A) \cdot \alpha^2 / \epsilon^2) | Hurt |
| Exploit nonuniform context frequencies with separate estimators for frequent and infrequent contexts | Robust MCB · Nonuniform context distributions | per-user interactions: O(\min(S,A)^{1-\gamma} \cdot \alpha / \epsilon^2); \tilde{O} (\alpha / \epsilon^2) | Helped |
Closest papers by meaning
- A Gang of Adversarial Bandits (NeurIPS 2021 · similarity 0.91)
- An Improved Algorithm for Adversarial Linear Contextual Bandits via Reduction (NeurIPS 2025 · similarity 0.88)
- Demystifying Online Clustering of Bandits: Enhanced Exploration Under Stochastic and Smoothed Adversarial Contexts (ICLR 2025 · similarity 0.88)
- Regret Bounds for Adversarial Contextual Bandits with General Function Approximation and Delayed Feedback (NeurIPS 2025 · similarity 0.88)
- Nearly Optimal Algorithms for Contextual Dueling Bandits from Adversarial Feedback (ICML 2025 · similarity 0.87)
- Faster Rates for Private Adversarial Bandits (ICML 2025 · 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.