Research map / Optimization / Nonconvex and smooth optimization
Momentum-Based Policy Gradient Methods
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 momentum-based variance reduction with importance sampling and an adaptive learning rate | IS-MBPG · Non-oblivious model-free RL | sample complexity: $O(\epsilon^{-4})$ -> $O(\epsilon^{-3})$ | Helped |
| Use Hessian-aided momentum-based variance reduction with an adaptive learning rate | HA-MBPG · Non-oblivious model-free RL | sample complexity: $O(\epsilon^{-4})$ -> $O(\epsilon^{-3})$ | Helped |
| Use importance-sampling momentum variance reduction with a non-adaptive learning rate | IS-MBPG* · Non-oblivious model-free RL | sample complexity: $O(\epsilon^{-4})$ -> $O(\epsilon^{-3})$ | Helped |
| Use one trajectory per iteration instead of large batches | IS-MBPG and HA-MBPG · Non-oblivious model-free RL | batch size: $O(\epsilon^{-2})$ -> $O(1)$ | Helped |
| Use IS-MBPG instead of HAPG or SRVR-PG | Policy gradient methods · CartPole | average episode return | Helped |
| Use IS-MBPG instead of the baselines | Policy gradient methods · Walker | average episode return | Helped |
| Use HA-MBPG instead of IS-MBPG | Momentum-based policy gradient · Walker | average episode return | Mixed |
| Use IS-MBPG or HA-MBPG instead of the baselines | Policy gradient methods · Hopper | learning speed | Helped |
| Use IS-MBPG or HA-MBPG instead of the baselines | Policy gradient methods · Hopper | final average reward | No clear effect |
| Use IS-MBPG instead of HAPG | Policy gradient methods · HalfCheetah | average episode return | Helped |
| Replace the exact Hessian-vector product with its finite-difference estimate | HA-MBPG · Four benchmark environments | average episode return | Hurt |
| Reduce the batch size to 20% of the original (ablation) | IS-MBPG and HA-MBPG · CartPole | average episode return: 20% | Helped |
Closest papers by meaning
- MDPGT: Momentum-Based Decentralized Policy Gradient Tracking (AAAI 2022 · similarity 0.91)
- Hessian Aided Policy Gradient (ICML 2019 · similarity 0.90)
- Sample Efficient Policy Gradient Methods with Recursive Variance Reduction (ICLR 2020 · similarity 0.89)
- Accelerated Policy Gradient: On the Convergence Rates of the Nesterov Momentum for Reinforcement Learning (ICML 2024 · similarity 0.88)
- Consensus Based Stochastic Optimal Control (ICML 2025 · similarity 0.88)
- Natural Policy Gradient as Doubly Smoothed Policy Iteration: A Bellman-Operator Framework (NeurIPS 2026 · 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.