Research map / Bandits and online learning / RL theory and MDPs
Learning with Good Feature Representations in Bandits and in RL with a Generative Model
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 a near-optimal experimental design and least-squares to identify an action | Action identification with d-dimensional features · Action identification with features | query complexity: c^{\max}_\delta(\mathcal{H}_\Phi^\varepsilon) \leq 4 d \log \log d + 16 | Helped |
| Relax the target error from near ε to order ε√d | Action identification with d-dimensional features · Action identification with features | query complexity: 4 d \log \log d + 16 | Helped |
| Require error tolerance close to the approximation error | Worst-case feature matrices · Action identification with features | query complexity: c_1^{\max}(\mathcal{H}_\Phi^\varepsilon) \geq (k+1)/2 | Hurt |
| Use phased elimination with near-optimal designs | Misspecified linear bandits · Misspecified linear bandit | regret: C\left[\sqrt{d n \log(nk)} + \varepsilon n \sqrt{d} \log(n)\right] | Helped |
| Use approximate policy iteration with design-core rollouts and feature-based least squares | RL with a generative model · Discounted RL with a generative model | maximum value suboptimality: C\varepsilon \sqrt{d} / (1 - \gamma)^2 | Helped |
| Add the misspecification-aware bonus to LinUCB | LinUCB · Linear contextual bandit | regret: O\left(d \sqrt{n} \log(n) + n \varepsilon \sqrt{d \log(n)}\right) | Helped |
| Use unmodified LinUCB under contextual misspecification | LinUCB · Linear contextual bandit constructed example | regret: at least n/2 | Hurt |
Closest papers by meaning
- Linear Bandits with Partially Observable Features (ICML 2025 · similarity 0.89)
- Sparsity-Agnostic Linear Bandits with Adaptive Adversaries (NeurIPS 2024 · similarity 0.88)
- Feasible Action Search for Bandit Linear Programs via Thompson Sampling (ICML 2025 · similarity 0.88)
- Causal Bandits: The Pareto Optimal Frontier of Adaptivity, a Reduction to Linear Bandits, and Limitations around Unknown Marginals (ICML 2024 · similarity 0.87)
- Generalized Linear Bandits with Limited Adaptivity (NeurIPS 2024 · similarity 0.87)
- Exploration via Feature Perturbation in Contextual Bandits (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.