Research map / Reinforcement learning / Visual and representation RL
When Is Generalizable Reinforcement Learning Tractable?
Dhruv Malik, Yuanzhi Li, Pradeep Kumar Ravikumar
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 |
|---|---|---|---|
| Rely on Weak Proximity, even with SIO, instead of a stronger shared-policy condition | Average Performance generalization · Average Performance | query cost: Ω \left( \min \left \{ \vert \ActSet \vert^H, 2^d \right\} \right) | Hurt |
| Rely on Weak Proximity, even with SIO, instead of a stronger shared-policy condition | Meta RL generalization · Meta RL | training query cost: Ω \left( \min \left \{ \vert \ActSet \vert^H, 2^d \right\} \right) | Hurt |
| Allow transition variation Θ(1/H) while keeping rewards shared under Weak Proximity and SIO | Average Performance generalization · Average Performance | query cost: Ω \left( \min \left \{ \vert \ActSet \vert^H, 2^d \right\} \right) | Hurt |
| Allow transition variation Θ(1/H) while keeping rewards shared under Weak Proximity and SIO | Meta RL generalization · Meta RL | training query cost: Ω \left( \min \left \{ \vert \ActSet \vert^H, 2^d \right\} \right) | Hurt |
| Use Strong Proximity with a policy near-optimal from every state, shared deterministic transitions, and WIO | GenRL · Average Performance | expected value suboptimality: \epsilon + 3 \polsub H + 3 \solvesub H | Improved |
| Keep both reward and transition variation o(1/H) | Generalization under the Simulation Lemma · Average Performance and Meta RL | value suboptimality: o(1) | Improved |
Closest papers by meaning
- On the Power of Pre-training for Generalization in RL: Provable Benefits and Hardness (ICML 2023 · similarity 0.89)
- On the Generalization Gap in Reparameterizable Reinforcement Learning (ICML 2019 · similarity 0.88)
- Instance-based Generalization in Reinforcement Learning (NeurIPS 2020 · similarity 0.88)
- On Generalization Across Environments In Multi-Objective Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Learning Generalizable and Composable Abstractions for Transfer in Reinforcement Learning (AAAI 2024 · similarity 0.88)
- Horizon Generalization in Reinforcement Learning (ICLR 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.
