Research map / Reinforcement learning / Policy gradient and actor-critic
Transfer of Samples in Policy Search via Multiple Importance Sampling
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 |
|---|---|---|---|
| Add multiple importance sampling (MIS) estimator to transfer source trajectories instead of plain per-decision IS | policy gradient transfer · LQR (transfer) | distance to optimal target parameter | Helped |
| Use per-decision IS (PD-IS) instead of GPOMDP for transfer | policy gradient transfer · LQR (transfer) | distance to optimal target parameter | Hurt |
| Add regression-based control variates to MIS estimator (ablation) | MIS / PD-MIS gradient estimator · LQR (transfer) | distance to optimal target parameter | Helped |
| Use per-decision MIS estimator instead of standard MIS for sample reuse (ablation) | standard policy gradient sample reuse · LQR (sample reuse) | learning speed vs GPOMDP | Helped |
| Transfer source samples with all transfer algorithms vs sample reuse of only target trajectories | policy search with model estimation · Cartpole | expected return | Helped |
| Estimate transition models with GP predictions plugged in directly as weights (ablation) | transfer with unknown models · Minigolf | expected return | Hurt |
| Use MSE-aware model estimation (discrete or RKHS) instead of ideal known models | transfer with unknown models · Cartpole and Minigolf | expected return | No clear effect |
| Plug GP-estimated models into planning (RL on GP model) instead of using them only for importance weights (ablation) | model-based policy learning · Cartpole | expected return | Hurt |
Closest papers by meaning
- Importance Weighted Transfer of Samples in Reinforcement Learning (ICML 2018 · similarity 0.92)
- Prioritizing Samples in Reinforcement Learning with Reducible Loss (NeurIPS 2023 · similarity 0.90)
- Demystifying the Paradox of Importance Sampling with an Estimated History-Dependent Behavior Policy in Off-Policy Evaluation (ICML 2025 · similarity 0.88)
- Generalizable Policy Improvement via Reinforcement Sampling (Student Abstract) (AAAI 2024 · similarity 0.87)
- Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems (NeurIPS 2025 · similarity 0.87)
- Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data (NeurIPS 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.