Research map / Reinforcement learning / Policy gradient and actor-critic
Policy Optimization via Importance Sampling
Alberto Maria Metelli, Matteo Papini, Francesco Faccio, Marcello Restelli
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 |
|---|---|---|---|
| Use A-POIS instead of TRPO | deep neural policies · Cart-Pole Balancing, Mountain Car, Double Inverted Pendulum, and Swimmer | average return: 4842.8 ± 13.0 vs 4869.8 ± 37.6; -63.7 ± 0.5 vs -61.7 ± 0.9; 4232.1 ± 189.5 vs 4412.4 ± 50.4; 88.7 ± 0.55 vs 96.0 ± 0.2 | Mixed |
| Use A-POIS instead of DDPG | deep neural policies · Cart-Pole Balancing, Mountain Car, Double Inverted Pendulum, and Swimmer | average return: 4634.4 ± 87.6 vs 4842.8 ± 13.0; -288.4 ± 170.3 vs -63.7 ± 0.5; 2863.4 ± 154.0 vs 4232.1 ± 189.5; 85.8 ± 1.8 vs 88.7 ± 0. | Improved |
| Use P-POIS instead of CEM | deep neural policies · Cart-Pole Balancing, Mountain Car, Double Inverted Pendulum, and Swimmer | average return: 4815.4 ± 4.8 vs 4428.1 ± 138.6; -66.0 ± 2.4 vs -78.9 ± 2.5; 2566.2 ± 178.9 vs 3161.4 ± 959.2; 68.8 ± 2.4 vs 76.8 ± 1.6 | Mixed |
| Use P-POIS instead of A-POIS | linear policies · Cartpole, Inverted Double Pendulum, Acrobot, and Mountain Car | average return | Improved |
| Use either POIS version instead of TRPO and PPO | linear policies · Inverted Pendulum | average return | Hurt |
| Set the A-POIS confidence parameter δ to an intermediate value rather than a small or near-one value | A-POIS · Cartpole | average return: 0.4 | Improved |
Closest papers by meaning
- Importance Sampling Techniques for Policy Optimization (JMLR 2020 · similarity 0.94)
- Optimistic Policy Optimization via Multiple Importance Sampling (ICML 2019 · similarity 0.91)
- Trajectory-Based Off-Policy Deep Reinforcement Learning (ICML 2019 · similarity 0.90)
- Low-Switching Policy Gradient with Exploration via Online Sensitivity Sampling (ICML 2023 · similarity 0.90)
- Ranking Policy Gradient (ICLR 2020 · similarity 0.88)
- Kernel Metric Learning for In-Sample Off-Policy Evaluation of Deterministic RL Policies (ICLR 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.
