Research map / Bandits and online learning / Multi-armed bandits
Conservative 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 |
|---|---|---|---|
| Replace BudgetFirst's initial fixed default-arm phase with confidence-based switching between exploration and default-arm play | Stochastic bandits · Stochastic bandits | pseudo-regret: BudgetFirst: \Omega\left(\frac{1}{\mu_0 \alpha} \sqrt{Kn \log\left(\frac{\log(n)}{\delta}\right)}\right); Conservative U | Helped |
| Add uniform-in-time return protection to stochastic UCB | Conservative UCB · Stochastic bandits | pseudo-regret: O(\sqrt{nKL}) -> O\paren*{\sqrt{nKL}+\frac{KL}{\alpha\mu_0}} | Hurt |
| Estimate rather than assume the default-arm mean is known | Conservative UCB · Stochastic bandits, unknown default-arm mean | pseudo-regret upper bound: 6L -> 7L | Hurt |
| Enforce the return constraint with safe play in adversarial bandits | Adversarial bandit algorithm · Adversarial bandits | constraint penalty in regret: O(\frac{KL}{\alpha\mu_0}) -> O\paren*{\frac{KL^2}{{(\alpha\mu_0)}^2}} | Hurt |
| Require the return constraint only at the final round rather than uniformly over time | Unbalanced MOSS · Stochastic bandits | early-round constraint satisfaction | Hurt |
| Use Conservative UCB without knowing the default-arm mean | Conservative UCB · Synthetic bandit environments | average regret | Hurt |
| Use knowledge of the default-arm mean | Conservative UCB · Synthetic bandit environments | average regret | Helped |
| Use the lower-confidence-bound-best arm instead of the default arm when the proposed UCB arm is unsafe (ablation) | Conservative UCB · Stochastic bandits | regret | Helped |
Closest papers by meaning
- Beyond Primal-Dual Methods in Bandits with Stochastic and Adversarial Constraints (NeurIPS 2024 · similarity 0.91)
- An Algorithm for Stochastic and Adversarial Bandits with Switching Costs (ICML 2021 · similarity 0.90)
- Stochastic Bandits Robust to Adversarial Attacks (ICLR 2025 · similarity 0.89)
- Taming Heavy-Tailed Losses in Adversarial Bandits and the Best-of-Both-Worlds Setting (NeurIPS 2024 · similarity 0.89)
- Improved Best-of-Both-Worlds Regret for Bandits with Delayed Feedback (NeurIPS 2025 · similarity 0.89)
- Regret Bounds for Adversarial Contextual Bandits with General Function Approximation and Delayed Feedback (NeurIPS 2025 · similarity 0.88)
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.