Research map / Bandits and online learning / Multi-armed bandits
Stochastic bandits with arm-dependent delays
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 a delay-corrected UCB with an additional tail-dependent bias term | PatientBandits · Stochastic delayed bandits | problem-dependent regret | Helped |
| Allow arm-dependent, potentially heavy-tailed delays and use PatientBandits | Stochastic delayed bandits · Problem-independent regret | regret | Mixed |
| Use PatientBandits rather than D-UCB when delay distributions vary across arms | Two-arm bandit with μ=(0.6,0.8), α₁=1 and α₂=0.3 · Heterogeneous Pareto delays | regret | Helped |
| Set the PatientBandits tail parameter below the true minimum tail index (ablation) | PatientBandits · Two-arm bandit, μ=(0.5,0.55), α₁=1 and α₂=0.3 | regret | Hurt |
| Set the PatientBandits tail parameter above the true minimum tail index (ablation) | PatientBandits · Two-arm bandit, μ=(0.5,0.55), α₁=1 and α₂=0.3 | regret | Hurt |
| Decrease the best arm's delay tail index (ablation) | PatientBandits · Two-arm bandits with varying gap Δ | regret | Hurt |
| Use well-calibrated parameters, ᾱ=0.5 and m=50 (ablation) | PatientBandits and D-UCB · Homogeneous delays, μ=(0.6,0.8), α₁=α₂=0.7 | regret | No clear effect |
| Increase D-UCB's threshold window (ablation) | D-UCB · Homogeneous delays, μ=(0.6,0.8), α₁=α₂=0.7 | early-stage regret | Hurt |
| Replace a fixed tail parameter with an online lower-confidence estimate of α under the stronger tail assumption | Adapt-PatientBandits · Stochastic delayed bandits | regret | Helped |
| Try to adapt to the delay-tail parameter under only the weak tail assumption | Any algorithm optimal for a fixed α · Two-arm stochastic delayed bandits | problem-independent regret | Hurt |
| Use a time-varying parameter αₜ=log log(t)/log(t) instead of knowing the fixed tail parameter (ablation) | PatientBandits · Fixed problem, asymptotic regime | problem-dependent regret | Helped |
Closest papers by meaning
- Near-Optimal Stochastic Linear Bandits with Delay (NeurIPS 2026 · similarity 0.91)
- Contextual Linear Bandits with Delay as Payoff (ICML 2025 · similarity 0.89)
- Delay as Payoff in MAB (AAAI 2025 · similarity 0.88)
- Best Arm Identification for Stochastic Rising Bandits (ICML 2024 · similarity 0.88)
- A Best-of-both-worlds Algorithm for Bandits with Delayed Feedback with Robustness to Excessive Delays (NeurIPS 2024 · similarity 0.88)
- Improved Regret Bounds for Linear Bandits with Heavy-Tailed Rewards (NeurIPS 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.