Research map / Bandits and online learning / Multi-armed bandits
Maximum Average Randomly Sampled: A Scale Free and Non-parametric Algorithm for Stochastic Bandits
Masoud Moravej Khorasani, Erik Weyer
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 MARS (random sub-sampled mean UCB) instead of Vanilla UCB when the assumed variance is wrong (set to 2) | MARS · Gaussian bandit, K=5, misspecified variance | cumulative regret | Improved |
| Use MARS instead of Bootstrapped UCB and GIRO under misspecified variance | MARS · Gaussian bandit, K=5, misspecified variance | cumulative regret | Improved |
| Use MARS instead of Vanilla UCB when variances are correctly known | MARS · Gaussian bandit, K=5, true variance | cumulative regret | Hurt |
| Use MARS instead of Normal-TS when the Gaussian model is correctly specified | MARS · Gaussian bandit, K=5, true variance | cumulative regret | Hurt |
| Use MARS instead of Vanilla UCB, Thompson sampling, Bootstrapped UCB and GIRO on uniform rewards | MARS · Uniform[-1,1] bandit, K=5 | cumulative regret | Improved |
| Raise PHE perturbation parameter a from 2.1 to 5.1 | PHE · Gaussian bandit, K=5, true variance | cumulative regret | Improved |
| Raise PHE parameter a from 2.1 to 5.1 | PHE · Uniform[-1,1] bandit, K=5 | cumulative regret | Hurt |
| Use MARS 95% bound instead of naive bootstrap, bootstrapped threshold, sub-Gaussian and Hoeffding bounds | MARS UCB · 95% UCB of sample mean, Gaussian and uniform data | upper confidence bound tightness | Improved |
| Compare MARS bound with naive bootstrap at small sample sizes | MARS UCB · 95% UCB of sample mean, small samples | upper confidence bound tightness | Hurt |
Closest papers by meaning
- Bootstrapping Upper Confidence Bound (NeurIPS 2019 · similarity 0.93)
- A Closer Look at the Worst-case Behavior of Multi-armed Bandit Algorithms (NeurIPS 2021 · similarity 0.91)
- Precise Asymptotics and Refined Regret of Variance-Aware UCB (NeurIPS 2025 · similarity 0.91)
- Noise-Adaptive Confidence Sets for Linear Bandits and Application to Bayesian Optimization (ICML 2024 · similarity 0.90)
- Gaussian Process Upper Confidence Bound Achieves Nearly-Optimal Regret in Noise-Free Gaussian Process Bandits (NeurIPS 2025 · similarity 0.90)
- Bayesian Regret Minimization in Offline Bandits (ICML 2024 · 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.
