Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Conservative Contextual 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 |
|---|---|---|---|
| Add a safety check that permits optimistic actions only when the worst-case confidence-set estimate satisfies the performance constraint | LUCB, yielding CLUCB · Linear contextual bandits (simulation) | safety constraint violations | Helped |
| Add conservative fallback to the baseline policy | LUCB, yielding CLUCB · Linear contextual bandits (theory) | regret: O\left(d\log\big(\frac{DT}{\lambda\delta}\big)\sqrt{T} + \frac{K\Delta_h}{\alpha r_l(\alpha r_l + \Delta_l)}\right) | Mixed |
| Use CLUCB2 to account for unknown baseline rewards with confidence bounds and observations from both action types | CLUCB · Linear contextual bandits (unknown baseline) | regret rate: O\left(d\log\left(\frac{DT}{\lambda\delta}\right)\sqrt{T}\right) | No clear effect |
| Use a tighter safety threshold, i.e. smaller α | CLUCB · Linear contextual bandits | regret | Hurt |
| Increase α, relaxing the safety constraint | CLUCB · Linear contextual bandits (simulation) | length of conservative phase | Helped |
| Use the baseline policy as the fallback during early exploration | CLUCB · Linear contextual bandits (simulation) | early per-step regret | Helped |
| Replace known baseline rewards with unknown rewards handled by confidence bounds (ablation) | CLUCB2 · Linear contextual bandits (theory) | number of conservative rounds: 256 \frac{d^2(B\sqrt{\lambda}+\sigma)^2}{\alpha^2 r_l^2}\left[\log\left(\frac{10d(B\sqrt{\lambda}+\sigma)\sqrt{D}}{\alph | Mixed |
| Update confidence sets using only optimistic-action observations (ablation) | CLUCB · Linear contextual bandits (known baseline) | confidence-set information | No clear effect |
| Use nested confidence sets (ablation) | CLUCB2 · Linear contextual bandits (unknown baseline) | computational complexity | Hurt |
Closest papers by meaning
- Conservative Contextual Bandits: Beyond Linear Representations (ICLR 2025 · similarity 0.94)
- Improved Algorithms for Conservative Exploration in Bandits (AAAI 2020 · similarity 0.92)
- Stage-wise Conservative Linear Bandits (NeurIPS 2020 · similarity 0.91)
- Learning in Generalized Linear Contextual Bandits with Stochastic Delays (NeurIPS 2019 · similarity 0.90)
- Strategic Linear Contextual Bandits (NeurIPS 2024 · similarity 0.88)
- Contextual Bandits for Unbounded Context Distributions (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.