Research map / Reinforcement learning / Visual and representation RL
On the Power of Pre-training for Generalization in RL: Provable Benefits and Hardness
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 |
|---|---|---|---|
| Pre-training on sampled training MDPs (PCE policy-value set) to reduce test-time fine-tuning regret, versus no pre-training | PCE test-stage fine-tuning · Test-time regret in K episodes (theoretical) | regret upper bound | Helped |
| Knowing the true distribution D in the asymptotic regime (K to infinity) | Test-time regret with prior knowledge of D · Asymptotic bandit-style regret (theoretical) | regret ratio lower bound vs. no-knowledge algorithm: liminf ratio >= 1/16 (constant-factor improvement at most) | No clear effect |
| Pre-training without test-time interaction via OMERM (optimistic model-based ERM over sampled training MDPs) | Expected-optimality policy learning · Sample complexity in training MDPs (theoretical) | number of sampled MDP instances for epsilon-optimal in expectation | Helped |
| Test-time interaction removed (deploy pre-trained policy without fine-tuning) | Policy optimality in instance · Proposition on instance-optimality (theoretical) | expected suboptimality gap: gap >= epsilon_0 (e.g., (M-1)/M for uniform D) | Hurt |
Closest papers by meaning
- Doubly Mild Generalization for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- How Ensembles of Distilled Policies Improve Generalisation in Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Provable Zero-Shot Generalization in Offline Reinforcement Learning (ICML 2025 · similarity 0.89)
- On Generalization Across Environments In Multi-Objective Reinforcement Learning (ICLR 2025 · similarity 0.88)
- The Generalization Gap in Offline Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning (ICML 2024 · 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.