Research map / Optimization / Nonconvex and smooth optimization
Policy Optimization with Stochastic Mirror Descent
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 the recursive variance-reduced policy-gradient estimator | VRMPO · epsilon-FOSP sample complexity | sample trajectories: O(ε^{-4}) -> O(ε^{-3}) | Helped |
| Use the recursive variance-reduced policy-gradient estimator | VRMPO · HalfCheetah | max-average return: 11781.07 -> 16095.51 | Helped |
| Use the recursive variance-reduced policy-gradient estimator | VRMPO · Hopper | max-average return: 3482.06 -> 3751.43 | Helped |
| Use the recursive variance-reduced policy-gradient estimator | VRMPO · Walker2d | max-average return: 5795.13 -> 5251.83 | Hurt |
| Use the recursive variance-reduced policy-gradient estimator | VRMPO · Reacher | max-average return: -0.44 -> -0.49 | Hurt |
| Use the recursive variance-reduced policy-gradient estimator | VRMPO · InvPendulum | max-average return: 9248.27 -> 9359.82 | Helped |
| Use a non-Euclidean mirror map (p ≠ 2) instead of the Euclidean map (p = 2) (ablation) | VRMPO · MuJoCo continuous control tasks | performance | Helped |
| Use the recursive variance-reduced estimator instead of the current-trajectory gradient estimator | MPO · SASC | convergence speed | Helped |
| Use VRMPO instead of REINFORCE | policy optimization · mirror-map experiment | convergence speed | Helped |
| Use VRMPO instead of SRVR-PG | policy optimization · mirror-map experiment | convergence speed | Helped |
| Use the shared trajectory for both gradient terms in the recursive estimator | VRMPO · policy-gradient estimation | gradient-estimator variance | Helped |
Closest papers by meaning
- Sample Efficient Policy Gradient Methods with Recursive Variance Reduction (ICLR 2020 · similarity 0.91)
- Learning Optimal Deterministic Policies with Stochastic Policy Gradients (ICML 2024 · similarity 0.88)
- Learning mirror maps in policy mirror descent (ICLR 2025 · similarity 0.87)
- Policy Mirror Descent with Lookahead (NeurIPS 2024 · similarity 0.87)
- Convergence Analysis of Policy Gradient Methods with Dynamic Stochasticity (ICML 2025 · similarity 0.87)
- Sharp Convergence and Sample Complexity of Policy Mirror Descent for Average-Reward MDPs (NeurIPS 2026 · similarity 0.86)
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.