Research map / Bandits and online learning / Multi-armed bandits
Optimal Algorithms for Stochastic Multi-Armed Bandits with Heavy Tailed Rewards
Kyungjae Lee, Hongjun Yang, Sungbin Lim, Songhwai Oh
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 APE² perturbation exploration instead of robust UCB | Heavy-tailed stochastic MAB · Small reward gap, Δ=0.3 or 0.1 | time-averaged cumulative regret | Improved |
| Use APE² perturbation exploration instead of robust UCB | Heavy-tailed stochastic MAB · p=1.5, Δ=0.8 | time-averaged cumulative regret | No clear effect |
| Use DSEE's fixed uniform exploration schedule instead of adaptive exploration | Heavy-tailed stochastic MAB · Simulated MAB | time-averaged cumulative regret | Hurt |
| Use the p-robust estimator instead of truncated mean | Reward estimation with Pareto noise · p=1.5, 1.9, 1.1; λₑ=1.0 | absolute estimation error | Improved |
| Use the p-robust estimator instead of truncated mean | Reward estimation with Pareto noise · p=1.1, λₑ=0.1 | absolute estimation error | No clear effect |
| Use the p-robust estimator instead of median of means | Reward estimation with Pareto noise · Simulated estimator convergence | absolute estimation error | Improved |
| Use the proposed perturbation strategy despite its slightly slower estimator convergence | APE² with p=1.1, λₑ=0.1 · Heavy-tailed MAB | time-averaged cumulative regret | Improved |
| Use Gumbel or Exponential perturbations rather than the other tested perturbations | APE² · Simulated heavy-tailed MAB | time-averaged cumulative regret | Improved |
Closest papers by meaning
- Stochastic Multi-Armed Bandits with Unrestricted Delay Distributions (ICML 2021 · similarity 0.90)
- Anytime optimal algorithms in stochastic multi-armed bandits (ICML 2016 · similarity 0.90)
- Stochastic Multi-armed Bandits: Optimal Trade-off among Optimality, Consistency, and Tail Risk (NeurIPS 2023 · similarity 0.89)
- Stochastic Multi-Armed Bandits with Control Variates (NeurIPS 2021 · similarity 0.89)
- Improved Regret Bounds for Linear Bandits with Heavy-Tailed Rewards (NeurIPS 2025 · similarity 0.89)
- Forced Exploration in Bandit Problems (AAAI 2024 · 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.
