Research map / Bandits and online learning / Multi-armed bandits
Observe Before Play: Multi-Armed Bandit with Pre-Observations
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 |
|---|---|---|---|
| Use UCB-ranked pre-observation lists instead of random lists | OBP-UCB · Single-player synthetic and real channel traces | average cumulative reward: 5%; 6% | Helped |
| Allow strategic pre-observations rather than only one observation | OBP-UCB · Single-player synthetic and real channel traces | average cumulative reward: 102%; 76% | Helped |
| Use collision-free UCB-based observation lists | C-MP-OBP · Multi-player synthetic and real channel traces | average cumulative reward: 7%; 4% | Helped |
| Use distributed randomized coordination without communication | D-MP-OBP · Multi-player synthetic and real channel traces | average cumulative reward: 39%; 30% | Helped |
| Increase pre-observation cost τ | OBP-UCB compared with random · Single-player synthetic and real channel traces | reward improvement over random: 5% -> 34% -> 140%; 6% -> 47% -> 245% | Helped |
| Increase pre-observation cost τ | OBP-UCB compared with single-observation baselines · Single-player synthetic and real channel traces | reward improvement over single-observation baseline: 102% -> 92% -> 78%; 76% -> 71% -> 63% | Hurt |
| Increase mean arm rewards | OBP-UCB, C-MP-OBP, and D-MP-OBP compared with random · Synthetic reward distributions | reward gap over random | Helped |
| Increase mean arm rewards across the range | OBP-UCB compared with single-observation baseline · Synthetic reward distributions | reward gap over single-observation baseline | Mixed |
| Use distributed rather than centralized coordination | D-MP-OBP versus C-MP-OBP · Multi-player simulations | regret | Hurt |
| Learn arm rankings and observation order with UCB | OBP-UCB · Single-player stochastic Bernoulli MAB | regret: O(K^2 log T) | Helped |
| Allocate disjoint observation lists centrally | C-MP-OBP · Centralized multi-player stochastic Bernoulli MAB | regret relative to greedy target policy: O(K^2 L^2 log T) | Helped |
| Use randomization to learn collision-free observation lists without communication | D-MP-OBP · Distributed multi-player stochastic Bernoulli MAB | loss relative to collision-free greedy policies: logarithmic in T | Helped |
Closest papers by meaning
- Distributed Multi-Player Bandits - a Game of Thrones Approach (NeurIPS 2018 · similarity 0.91)
- Multiple-Play Stochastic Bandits with Shareable Finite-Capacity Arms (ICML 2022 · similarity 0.91)
- Strategic Multi-Armed Bandit Problems Under Debt-Free Reporting (NeurIPS 2024 · similarity 0.90)
- Multi-Armed Bandits With Best-Action Queries (NeurIPS 2026 · similarity 0.89)
- Fusing Reward and Dueling Feedback in Stochastic Bandits (ICML 2025 · similarity 0.89)
- Nash Social Welfare for Multi Armed Bandits: Trajectory-wise Expected and High Probability Regret (NeurIPS 2026 · 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.