Research map / Reinforcement learning / MDPs and planning
Bridging RL Theory and Practice with the Effective Horizon
Cassidy Laidlaw, Stuart Russell, Anca Dragan
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 |
|---|---|---|---|
| predict PPO sample complexity with effective horizon bound instead of EPW bound | sample complexity bounds · BRIDGE, PPO | Spearman correlation: 0.69 -> 0.81 | Improved |
| predict PPO sample complexity with effective horizon bound instead of covering length | sample complexity bounds · BRIDGE, PPO | Spearman correlation: 0.35 -> 0.81 | Improved |
| predict DQN sample complexity with effective horizon bound instead of EPW | sample complexity bounds · BRIDGE, DQN | Spearman correlation: 0.58 -> 0.74 | Improved |
| predict PPO convergence with effective horizon instead of worst-case bound | sample complexity bounds · BRIDGE, PPO | AUROC: 0.57 -> 0.92 | Improved |
| predict DQN convergence with effective horizon instead of EPW | sample complexity bounds · BRIDGE, DQN | accuracy: 0.85 -> 0.86 | Improved |
| use effective horizon instead of UCB bound for absolute tightness | sample complexity bounds · BRIDGE, PPO | median ratio: 20 -> 31 | Hurt |
| use effective horizon instead of EPW for absolute tightness | sample complexity bounds · BRIDGE, PPO | median ratio: $1.1 \times 10^{5}$ -> 31 | Improved |
| use effective horizon to predict effect of reward shaping instead of EPW | effective horizon bound · 77 shaped MiniGrid envs, PPO | correlation of predicted vs empirical change: 0.20 -> 0.48 | Improved |
| use effective horizon to predict effect of reward shaping instead of EPW | effective horizon bound · 77 shaped MiniGrid envs, DQN | correlation of predicted vs empirical change: 0.70 -> 0.35 | Hurt |
| use effective horizon instead of covering length to predict effect of pretrained-policy init | effective horizon bound · 82 MDPs, PPO | correlation of predicted vs empirical change: -0.36 -> 0.57 | Improved |
| predict PPO sample complexity with effective horizon bound instead of GORP empirical | empirical proxy · BRIDGE, PPO | Spearman correlation: 0.79 -> 0.81 | Improved |
| predict DQN sample complexity from PPO empirical results instead of effective horizon | empirical proxy · BRIDGE, DQN | Spearman correlation: 0.74 -> 0.77 | Improved |
Closest papers by meaning
- The Effective Horizon Explains Deep RL Performance in Stochastic Environments (ICLR 2024 · similarity 0.94)
- Horizon Reduction Makes RL Scalable (NeurIPS 2025 · similarity 0.89)
- The Reasoning Boundary Paradox: How Reinforcement Learning Constrains Language Models (NeurIPS 2026 · similarity 0.88)
- Model-based RL as a Minimalist Approach to Horizon-Free and Second-Order Bounds (ICLR 2025 · similarity 0.88)
- MATH-Beyond: A Benchmark for RL to Expand Beyond the Base Model (ICLR 2026 · similarity 0.88)
- Optimal Single-Policy Sample Complexity and Transient Coverage for Average-Reward Offline RL (NeurIPS 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.
