Research map / Bandits and online learning / Multi-armed bandits
Optimistic Gittins Indices
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 OGI(1) instead of IDS | OGI · Gaussian bandit | mean regret: 55.83 -> 49.19 | Helped |
| Use OGI(1) instead of Thompson sampling | OGI · Gaussian bandit | mean regret: 67.40 -> 49.19 | Helped |
| Use OGI(1) instead of Bayes UCB | OGI · Gaussian bandit | mean regret: 60.30 -> 49.19 | Helped |
| Use specialized Gaussian Gittins approximation in OGI (ablation) | OGI(1) · Gaussian bandit | mean regret: 49.19 -> 47.64 | Helped |
| Replace IDS with OGI(1) for compute | OGI · Gaussian bandit | CPU time per trial (s): 11.18 -> 0.02 | Helped |
| Look ahead 3 steps instead of 1 in OGI (ablation) | OGI · Bernoulli bandit | mean regret: 18.12 -> 18.00 | Helped |
| Use exact Gittins indices instead of 1-step optimistic approximation (ablation) | OGI · Bernoulli bandit | mean regret: 18.12 -> 17.52 | Helped |
| Use OGI(1) instead of IDS | OGI · Bernoulli bandit | mean regret: 19.03 -> 18.12 | Helped |
| Use OGI(1) instead of Thompson sampling | OGI · Bernoulli bandit | mean regret: 27.39 -> 18.12 | Helped |
| Increase lookahead K from 1 to 3 (ablation) | OGI · Bernoulli bandit | approximation error to Gittins index: around 15% -> around 4% | Helped |
| Increase lookahead K from 1 to 3 (ablation) | OGI · Bernoulli bandit | CPU time (s): 0.19 -> 0.89 | Hurt |
| Vary tuning parameter alpha over 50, 100, 150 (ablation) | OGI · Longer horizon 5000 steps, 3 arms | regret | No clear effect |
Closest papers by meaning
- Noise-Adaptive Confidence Sets for Linear Bandits and Application to Bayesian Optimization (ICML 2024 · similarity 0.88)
- The Choice of Noninformative Priors for Thompson Sampling in Multiparameter Bandit Models (AAAI 2024 · similarity 0.88)
- On Interpolating Experts and Multi-Armed Bandits (ICML 2024 · similarity 0.88)
- Bayesian Analysis of Combinatorial Gaussian Process Bandits (ICLR 2025 · similarity 0.87)
- Cost-aware Bayesian Optimization via the Pandora's Box Gittins Index (NeurIPS 2024 · similarity 0.87)
- Only Pay for What Is Uncertain: Variance-Adaptive Thompson Sampling (ICLR 2024 · 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.