Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Langevin Thompson Sampling with Logarithmic Communication: Bandits and Reinforcement Learning
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 |
|---|---|---|---|
| dynamic doubling batching (BLTS) instead of fully sequential updates for SGLD-TS (ablation) | SGLD-TS on Gaussian bandits · Gaussian Bandits (N=15) | regret: 99.80 vs 99.66 fully sequential | No clear effect |
| static batching instead of dynamic batching for SGLD-TS (ablation) | SGLD-TS on Gaussian bandits · Gaussian Bandits (N=15) | regret: 148.52 vs 99.80 dynamic | Hurt |
| replace exact TS with SGLD approximate sampling under dynamic batching | Thompson sampling on Gaussian bandits · Gaussian Bandits (N=15) | regret: 99.80 vs 98.71 | No clear effect |
| Langevin TS (SGLD-TS dynamic batch) instead of UCB1 on Gaussian bandits | SGLD-TS · Gaussian Bandits (N=15) | regret: 99.80 vs 153.31 | Helped |
| MLD sampling with static policy switches instead of TSDE dynamic switching | PSRL on RiverSwim · RiverSwim (|S|=5, |A|=2) | max policy switches: 12.0 vs 94.0 | Helped |
| dynamic batching of visit counts (DB-PSRL) instead of static doubling | PSRL on RiverSwim · RiverSwim (|S|=5, |A|=2) | average reward: 2.41 vs 4.01 for MLD-PSRL | Hurt |
| MLD sampling with static policy switches (MLD-PSRL) vs DS-PSRL | PSRL on RiverSwim · RiverSwim (|S|=5, |A|=2) | average reward: 4.01 vs 4.02 | No clear effect |
Closest papers by meaning
- On Approximate Thompson Sampling with Langevin Algorithms (ICML 2020 · similarity 0.92)
- No-Regret Thompson Sampling for Finite-Horizon Markov Decision Processes with Gaussian Processes (NeurIPS 2025 · similarity 0.91)
- Batched Thompson Sampling (NeurIPS 2021 · similarity 0.91)
- Provable and Practical: Efficient Exploration in Reinforcement Learning via Langevin Monte Carlo (ICLR 2024 · similarity 0.90)
- Feel-Good Thompson Sampling for Contextual Bandits: a Markov Chain Monte Carlo Showdown (NeurIPS 2025 · similarity 0.90)
- The Choice of Noninformative Priors for Thompson Sampling in Multiparameter Bandit Models (AAAI 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.