Research map / Optimization / Optimizers
Model-Based Reparameterization Policy Gradient Methods: Theory and Practical 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 |
|---|---|---|---|
| add spectral normalization to model and policy networks (ablation) | RP-DP (MAAC-style) · walker2d and hopper, MuJoCo, long unroll | return | Helped |
| add spectral normalization to model and policy networks (ablation) | RP-DP · walker2d h=10, hopper h=15 | gradient variance | Helped |
| add spectral normalization at small unroll length (ablation) | RP PGMs · MuJoCo locomotion | return | No clear effect |
| increase model unroll length h without SN (ablation) | vanilla RP-DP · hopper loss landscape | gradient variance and landscape smoothness | Hurt |
| increase model unroll length h with SN (ablation) | RP-DP-SN · walker2d and hopper | return | No clear effect |
| add SN to the learned model in differentiable simulator setting despite higher gradient bias (ablation) | RP-DP · dFlex locomotion | return | Helped |
| replace learned smooth proxy model with analytic policy gradient from simulator | RP-DP-SN · dFlex hopper and half-cheetah | sample efficiency | Hurt |
| compare model-based RP-DP against LR policy gradients (REINFORCE, NPG, ACKTR, PPO) and MBPO | RP-DP · MuJoCo tasks | return | Helped |
Closest papers by meaning
- Reparameterization Proximal Policy Optimization (ICML 2026 · similarity 0.90)
- Reparameterization Flow Policy Optimization (ICML 2026 · similarity 0.89)
- Reducing Reparameterization Gradient Variance (NeurIPS 2017 · similarity 0.87)
- Reusing Trajectories in Policy Gradients Enables Fast Convergence (ICML 2026 · similarity 0.87)
- Convergence Analysis of Policy Gradient Methods with Dynamic Stochasticity (ICML 2025 · similarity 0.86)
- Rao-Blackwellised Reparameterisation Gradients (NeurIPS 2025 · 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.