Research map / Bandits and online learning / Contextual bandits and Thompson sampling
PAC-Bayesian Offline Contextual Bandits With Guarantees
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 |
|---|---|---|---|
| Optimize the C_n bound instead of the LS_n bound | LIG policies · FashionMNIST, EMNIST-b, NUS-WIDE-128, Mediamill | guaranteed improvement over the logging policy | Helped |
| Optimize the CBB_n bound with ξ = -1/2 instead of LS_n | LIG policies · FashionMNIST, EMNIST-b, NUS-WIDE-128, Mediamill | guaranteed improvement over the logging policy | Helped |
| Use CBB_n with ξ = -1/2 rather than ξ = 0 (ablation) | CBB_n policy optimization · FashionMNIST, EMNIST-b, NUS-WIDE-128, Mediamill | guaranteed risk | Helped |
| Use CBB_n with ξ = 0 rather than C_n (ablation) | LIG policies · Uniform logging policies | guaranteed improvement | Hurt |
| Use CBB_n with ξ = -1/2 rather than C_n (ablation) | LIG policies · NUS-WIDE-128 with near-uniform logging policies | guaranteed improvement | Hurt |
| Use C_n rather than CBB_n with ξ = -1/2 (ablation) | LIG policies · FashionMNIST and EMNIST-b with peaked logging policies | guaranteed improvement | Hurt |
| Optimize LS_n rather than the proposed C_n or CBB_n bounds | LIG and Mixed Logit policies · FashionMNIST, EMNIST-b, NUS-WIDE-128, Mediamill | guaranteed improvement over the logging policy | Hurt |
| Use bounds with lower guaranteed risk to select the policy | LIG policies · FashionMNIST, EMNIST-b, NUS-WIDE-128, Mediamill | true risk | Helped |
Closest papers by meaning
- Logarithmic Smoothing for Pessimistic Off-Policy Evaluation, Selection and Learning (NeurIPS 2024 · similarity 0.90)
- Cost-Aware Near-Optimal Policy Learning (AAAI 2025 · similarity 0.89)
- Bayesian Regret Minimization in Offline Bandits (ICML 2024 · similarity 0.89)
- Offline Oracle-Efficient Learning for Contextual MDPs via Layerwise Exploration-Exploitation Tradeoff (NeurIPS 2024 · similarity 0.89)
- Leveraging (Biased) Information: Multi-armed Bandits with Offline Data (ICML 2024 · similarity 0.89)
- Selective Uncertainty Propagation in Offline RL (AAAI 2025 · similarity 0.88)
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.