Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Profile-Based Bandit with Unknown Profiles
Sylvain Lamprier, Thibault Gisselbrecht, Patrick Gallinari
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 |
|---|---|---|---|
| add the profile exploration term (confidence ball on profiles) instead of using empirical sample means as true profiles (ablation) | SampLinUCB p=0 · Twitter data capture (USElections, OlympicGames, Brexit) | cumulative reward | Improved |
| add the profile exploration term (confidence ball on profiles) (ablation) | SampLinUCB p=0 · Artificial data (K=100, d=5, T=30000) | cumulative pseudo-regret | Improved |
| use profile-based SampLinUCB p=0 instead of the best non-contextual policy | Twitter data capture · Twitter data capture (non-Sport reward functions) | final cumulative reward: between 1.5 and 3 times the best non-contextual approach | Improved |
| use SampLinUCB p=0 instead of CUCBV on the Sport reward function | Twitter data capture · Twitter data capture (Sport reward) | cumulative reward | No clear effect |
| deliver profile samples for all arms (p=1) instead of only selected arms (p=0) (ablation) | SampLinUCB · Twitter data capture | final cumulative reward: p=0 reaches at least 80% of the reward of p=1 | Improved |
| observe profile samples at random for small p (0.005, 0.01) instead of only for selected arms (p=0) (ablation) | SampLinUCB · Artificial data (sigma=2.0 and 1.0) | cumulative pseudo-regret | Hurt |
| use profile-conditioned LinUCB on current samples instead of SampLinUCB | LinUCB baseline · Artificial data and Twitter data capture | cumulative regret / reward | Hurt |
| lower profile sample variance sigma (2.0 to 0.5) (ablation) | SampLinUCB p=0 · Artificial data | cumulative pseudo-regret | Improved |
Closest papers by meaning
- Feature and Parameter Selection in Stochastic Linear Bandits (ICML 2022 · similarity 0.87)
- Combinatorial Stochastic-Greedy Bandit (AAAI 2024 · similarity 0.87)
- Contextual Bandits with Large Action Spaces: Made Practical (ICML 2022 · similarity 0.86)
- Stochastic Bandits with Context Distributions (NeurIPS 2019 · similarity 0.86)
- Demystifying Online Clustering of Bandits: Enhanced Exploration Under Stochastic and Smoothed Adversarial Contexts (ICLR 2025 · similarity 0.86)
- Small steps no more: Global convergence of stochastic gradient bandits for arbitrary learning rates (NeurIPS 2024 · similarity 0.86)
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.
