Research map / Bandits and online learning / Multi-armed bandits
Strategies for Safe Multi-Armed Bandits with Logarithmic Regret and Risk
Tianrui Chen, Aditya Gangrade, Venkatesh Saligrama
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 single-arm doubly optimistic index schema instead of policy-level (simplex) methods | safe bandit with single optimal arm · 4-arm Bernoulli instance, alpha=0.5, single optimal arm 3 (mu3=0.6) | unsafe arm plays: about 550 vs at least 8000 | Improved |
| Use policy-level (BwCR, Pess) methods rather than per-arm index schema | safe bandit with multiple optimal policies · 4-arm instance, alpha=0.5 | expected unsafe plays E[U_t] | Hurt |
| Replace Thompson sampling selection with KL-UCB-based index in the permissible-set scheme (TS with optimistic safety indices vs KL-UCB) | drug trial data alpha=0.21 · Drug trial (Phase 2 RA data), alpha=0.21 | regret and unsafe plays | Hurt |
| Use BayesUCB safety index with Thompson sampling (TS+BayesUCB) instead of KL-UCB confidence bounds | safe bandit on drug trial data · Drug trial (Phase 2 RA data), alpha=0.21 | regret and unsafe plays | Improved |
| Naive Thompson sampling on safety posterior compared against fixed alpha (sigma sampled from Beta, Pi = {sigma<=alpha}) | safe bandit with nu*=alpha · simulation with nu*=alpha | expected regret: linear expected regret | Hurt |
| Uniform quantile schedule delta=1/(t+1) instead of conservative delta_t^k=(sqrt(8N) t log^3 t)^-1 in TS+BayesUCB | TS+BayesUCB · simulations | regret | Improved |
Closest papers by meaning
- Safe Linear Stochastic Bandits (AAAI 2020 · similarity 0.92)
- Stochastic Multi-armed Bandits: Optimal Trade-off among Optimality, Consistency, and Tail Risk (NeurIPS 2023 · similarity 0.91)
- Near-Optimal MNL Bandits Under Risk Criteria (AAAI 2021 · similarity 0.91)
- Bayesian Regret Minimization in Offline Bandits (ICML 2024 · similarity 0.90)
- Nash Social Welfare for Multi Armed Bandits: Trajectory-wise Expected and High Probability Regret (NeurIPS 2026 · similarity 0.89)
- Causal Bandits: The Pareto Optimal Frontier of Adaptivity, a Reduction to Linear Bandits, and Limitations around Unknown Marginals (ICML 2024 · similarity 0.89)
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.
