Research map / Bandits and online learning / Multi-armed bandits
Linear Stochastic Bandits Under Safety Constraints
Sanae Amani, Mahnoosh Alizadeh, Christos Thrampoulidis
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 |
|---|---|---|---|
| Use a pure-exploration phase before safe exploration-exploitation (ablation) | Safe-LUCB · Linear stochastic bandit | cumulative regret | Improved |
| Choose the pure-exploration duration using known positive safety gap Δ | Safe-LUCB · Linear stochastic bandit, Δ>0 | cumulative regret: \Otilde(\sqrt{T}) versus \Otilde(T^{2/3}) | Improved |
| Set the pure-exploration duration for the worst case, without using knowledge of Δ (ablation) | Safe-LUCB · Linear stochastic bandit, Δ>0 | cumulative regret: \Otilde(T^{2/3}) versus \Otilde(\sqrt{T}) | Hurt |
| Adapt the pure-exploration duration using a lower confidence bound on Δ | Safe-LUCB · K-armed bandit with strictly positive safety gap | average per-step regret | Improved |
| Replace the standard confidence region with the modified ℓ1-confidence region (ablation) | Safe-LUCB · Linear stochastic bandit | cumulative regret: an extra factor of \sqrt{d} | Hurt |
| Replace the standard confidence region with the modified ℓ1-confidence region (ablation) | Safe-LUCB · Linear stochastic bandit | computational tractability | Improved |
| Use nested confidence regions (ablation) | Safe-LUCB · Linear stochastic bandit, Δ>0 | pure-exploration duration: T' becomes the smallest value satisfying ... which is now a constant independent of T | Improved |
Closest papers by meaning
- Safe Linear Stochastic Bandits (AAAI 2020 · similarity 0.94)
- Active Learning with Safety Constraints (NeurIPS 2022 · similarity 0.90)
- Feasible Action Search for Bandit Linear Programs via Thompson Sampling (ICML 2025 · similarity 0.90)
- Constrained Linear Thompson Sampling (NeurIPS 2025 · similarity 0.89)
- Conservative Contextual Bandits: Beyond Linear Representations (ICLR 2025 · similarity 0.88)
- Learning Safe Control via On-the-Fly Bandit Exploration (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.
