Research map / Reinforcement learning / Visual and representation RL
Evaluating the Performance of Reinforcement Learning Algorithms
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 complete algorithm definition with adaptive Parl2 step sizes instead of the standard Sarsa(λ) definition | Sarsa(λ) · 15 environments | aggregate performance: 0.0831 -> 0.4623 | Helped |
| Use a complete algorithm definition with adaptive Parl2 step sizes instead of the standard Q(λ) definition | Q(λ) · 15 environments | aggregate performance: 0.0640 -> 0.4366 | Helped |
| Use Sarsa(λ) rather than Q(λ) | λ-return control algorithms · 15 environments | aggregate performance: 0.0640 -> 0.0831 | Helped |
| Use AC rather than NAC-TD | actor-critic algorithms · 15 environments | aggregate performance: 0.0516 -> 0.0785 | Helped |
| Use nonparametric PBP bounds rather than PBP-t bounds | aggregate confidence intervals · discrete environments and Mountain Car, 10 samples | failure rate: 1.000 -> 0.0 | Helped |
| Use bootstrap bounds rather than PBP bounds | aggregate confidence intervals · discrete environments and Mountain Car, 10,000 samples | proportion of significant pairwise comparisons: 0.33 -> 0.83 | Helped |
| Use bootstrap bounds rather than PBP bounds | aggregate confidence intervals · discrete environments and Mountain Car, 10,000 samples | failure rate: 0.0 -> 0.069 | Hurt |
| Use PBP-t bounds rather than bootstrap bounds | aggregate confidence intervals · discrete environments and Mountain Car, 10,000 samples | failure rate: 0.069 -> 0.003 | Helped |
Closest papers by meaning
- Position: Benchmarking is Limited in Reinforcement Learning Research (ICML 2024 · similarity 0.93)
- Beyond Expected Return: Accounting for Policy Reproducibility When Evaluating Reinforcement Learning Algorithms (AAAI 2024 · similarity 0.90)
- Doubly Optimal Policy Evaluation for Reinforcement Learning (ICLR 2025 · similarity 0.88)
- A Method for Evaluating Hyperparameter Sensitivity in Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- Efficient Multi-Policy Evaluation for Reinforcement Learning (AAAI 2025 · similarity 0.88)
- Efficient Policy Evaluation with Offline Data Informed Behavior Policy Design (ICML 2024 · 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.