Research map / Bandits and online learning / Multi-armed bandits
Unimodal Thompson Sampling for Graph-Structured Arms
Stefano Paladino, Francesco Trovò, Marcello Restelli, Nicola Gatti
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 |
|---|---|---|---|
| Restrict Thompson sampling to the leader and its neighbors, with periodic leader pulls | Thompson Sampling for unimodal graph-structured arms · Line graph, K=17 | regret ratio vs OSUB: 0.52 \pm 0.07 vs 1.34 \pm 0.07 for TS | Improved |
| Restrict Thompson sampling to the leader and its neighbors, with periodic leader pulls | Thompson Sampling for unimodal graph-structured arms · Line graph, K=129 | regret ratio vs OSUB: 0.76 \pm 0.15 vs 2.68 \pm 0.05 for TS | Improved |
| Exploit the unimodal graph structure using UTS rather than TS | UTS vs TS · Erdős–Rényi graphs, p=1 | pseudo-regret: 34 \pm 0.4 vs 18 \pm 0.2 (K=5, KLUCB vs TS); 17 \pm 0.1 vs 18 \pm 0.2 (K=5, UTS vs TS) | Improved |
| Use UTS instead of OSUB | Unimodal graph-structured bandits · Erdős–Rényi graphs, except K=1000 and p=ell | pseudo-regret | Improved |
| Use OSUB instead of UTS | Unimodal graph-structured bandits · Erdős–Rényi graph, K=1000, p=ell | pseudo-regret: 1165 \pm 20.7 vs 1165 \pm 41.8 | No clear effect |
| Exploit the graph structure with OSUB instead of general-purpose TS | OSUB vs TS · Erdős–Rényi graphs, p=0.1, K≥10 | pseudo-regret: 66 vs 64 (K=10); 162 vs 126 (K=20); 278 vs 144 (K=50); 519 vs 266 (K=100); 4564 vs 2358 (K=1000) | Improved |
| Use UTS rather than OSUB when neighboring rewards are extremely close | UTS vs OSUB · Line graphs with very small adjacent reward gaps | pseudo-regret: T=10^5; T=10^7; |μ_i - μ_{i+1}| ∈ {0.001, 0.002, 0.005} | Mixed |
| Use UTS rather than OSUB | UTS vs OSUB · Erdős–Rényi graphs with p=5/K or p=10/K | pseudo-regret: T=10^6 | Improved |
Closest papers by meaning
- Thompson Sampling for Stochastic Bandits with Graph Feedback (AAAI 2017 · similarity 0.91)
- Finite-Time Frequentist Regret Bounds of Multi-Agent Thompson Sampling on Sparse Hypergraphs (AAAI 2024 · similarity 0.89)
- Thompson Sampling for Real-Valued Combinatorial Pure Exploration of Multi-Armed Bandit (AAAI 2024 · similarity 0.88)
- Efficient Graph Bandit Learning with Side-Observations and Switching Constraints (AAAI 2025 · similarity 0.88)
- Thompson Sampling For Combinatorial Bandits: Polynomial Regret and Mismatched Sampling Paradox (NeurIPS 2024 · similarity 0.88)
- Feel-Good Thompson Sampling for Contextual Bandits: a Markov Chain Monte Carlo Showdown (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.
