Research map / Reinforcement learning / Policy gradient and actor-critic
Policy Optimization as Wasserstein Gradient Flows
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 WGF-DP-V instead of SAC | Direct policy optimization · Swimmer, Walker, Hopper, Humanoid | Episodes to threshold: 76 vs 112; 2289 vs 2388; 678 vs 736; 18740 vs 26476 | Helped |
| Use WGF-DP-V instead of TRPO-GAE | Direct policy optimization · Swimmer, Walker, Hopper, Humanoid | MaxReturn: 181.60 vs 110.58; 4978.59 vs 3497.81; 3248.76 vs 2604; 3077.84 vs 5411.15 | Mixed |
| Add the value-network variance-reduction variant | DP-WGF · MuJoCo tasks: Swimmer, Hopper, Walker, Humanoid | Average return | Helped |
| Make the Wasserstein trust-region scale excessively large | DP-WGF-V on Swimmer-v1 · Swimmer-v1 | Final reward | Hurt |
| Use a moving-average previous policy rather than the last-iteration policy | DP-WGF-V · Stable-learning settings | Convergence speed | Hurt |
| Use a moving-average previous policy rather than the last-iteration policy | DP-WGF-V · Unstable-learning settings | Training stability | Helped |
Closest papers by meaning
- Efficient Wasserstein Natural Gradients for Reinforcement Learning (ICLR 2021 · similarity 0.92)
- Wasserstein Policy Optimization (ICML 2025 · similarity 0.91)
- Learning to Score Behaviors for Guided Policy Optimization (ICML 2020 · similarity 0.89)
- On the Second-Order Convergence of Biased Policy Gradient Algorithms (ICML 2024 · similarity 0.88)
- Natural Policy Gradient as Doubly Smoothed Policy Iteration: A Bellman-Operator Framework (NeurIPS 2026 · similarity 0.87)
- Convergence Analysis of Policy Gradient Methods with Dynamic Stochasticity (ICML 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.