Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Misspecified Linear Bandits
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 RLB's linearity test to select OFUL or UCB | Bandit learning with perfectly linear rewards · Perfectly linear bandit | regret: O(d√T) for RLB; O(d√T) for OFUL | No clear effect |
| Use RLB's linearity test to switch to UCB on large non-sparse deviations | OFUL · Large non-sparse deviation | regret: Ω(T) for OFUL; O(√NT) for RLB | Helped |
| Apply a sparse reward perturbation to one arm | Any linear bandit algorithm with optimal linear-model regret · Large sparse deviation | expected regret: O(d√T) -> Ω(T) | Hurt |
| Use OFUL despite a large non-sparse deviation | OFUL · Large non-sparse deviation | expected regret: Ω(T) | Hurt |
| Allow small deviation from linear rewards | OFUL · Small deviation | regret: O(d√T) | No clear effect |
| Use RLB on synthetic perfectly linear data | RLB · Synthetic data, H0 | regret: 0.0001 false-alarm probability; 0.9999 probability of closely following OFUL | No clear effect |
| Use RLB to select UCB on synthetic non-sparse deviations | OFUL · Synthetic data, H1 with |ε_i| > 2 | regret: 0.0001 miss-detection error | Helped |
| Increase RLB sampling count to k=70 (ablation) | RLB · Yahoo! Learning to Rank set2.test.txt, N=500, d=50 | regret: k=70; 3570 rounds | Mixed |
| Reduce RLB sampling count to k=10 (ablation) | RLB · Yahoo! Learning to Rank set2.test.txt, N=500, d=50 | regret: k=10 | Hurt |
| Use a reasonable sampling count k=50 rather than k=10 (ablation) | RLB · Yahoo! Learning to Rank set2.test.txt, N=800, d=50 | regret: k=50; k=10 | Helped |
Closest papers by meaning
- Corruption-Robust Linear Bandits: Minimax Optimality and Gap-Dependent Misspecification (NeurIPS 2024 · similarity 0.91)
- Generalized Linear Bandits: Almost Optimal Regret with One-Pass Update (NeurIPS 2025 · similarity 0.91)
- Sparsity-Agnostic Linear Bandits with Adaptive Adversaries (NeurIPS 2024 · similarity 0.91)
- Bayesian Regret Minimization in Offline Bandits (ICML 2024 · similarity 0.90)
- Linear Bandits with Partially Observable Features (ICML 2025 · similarity 0.90)
- Near-Optimal Stochastic Linear Bandits with Delay (NeurIPS 2026 · similarity 0.90)
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.