Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Improved Algorithms for Conservative Exploration in 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 the confidence-ellipsoid lower bound on accumulated rewards with a Martingale lower bound (ablation) | CUCB · Synthetic Bernoulli MAB, worst-case instance | regret: 43% | Helped |
| Choose the most optimistic safe arm instead of taking the baseline when the UCB arm is unsafe (ablation) | CUCB · Synthetic Bernoulli MAB, worst-case instance | regret: 12% | Helped |
| Combine the Martingale lower bound and constrained action selection (ablation) | CUCB · Synthetic Bernoulli MAB, worst-case instance | regret: more than 51% | Helped |
| Use constrained action selection to play safe, optimistic non-UCB arms (ablation) | CLinUCB · Synthetic linear bandit, worst-case model | regret: 38% | Helped |
| Use the full conservative constrained algorithm | CLinUCB · Jester, 40-joke subset | regret: less than half of CLinUCB | Helped |
| Increase the interval between conservative checkpoints (ablation) | CCLinUCBc · Synthetic Bernoulli MAB | regret difference from UCB | Helped |
| Use checkpoint-based conservative verification rather than checking safety every step (ablation) | CCLinUCBc · Jester, 40-joke subset | regret | Helped |
| Replace exact conservative-condition evaluation with a statistical lower bound | CUCB · Synthetic Bernoulli MAB | regret | Hurt |
Closest papers by meaning
- Conservative Contextual Bandits: Beyond Linear Representations (ICLR 2025 · similarity 0.92)
- A One-Size-Fits-All Solution to Conservative Bandit Problems (AAAI 2021 · similarity 0.92)
- Stage-wise Conservative Linear Bandits (NeurIPS 2020 · similarity 0.92)
- Bayesian Regret Minimization in Offline Bandits (ICML 2024 · similarity 0.88)
- Adaptive Regret for Bandits Made Possible: Two Queries Suffice (ICLR 2024 · similarity 0.88)
- Non-stationary Bandit Convex Optimization: A Comprehensive Study (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.