Research map / Reinforcement learning / Policy gradient and actor-critic
Online Meta-Critic Learning for Off-Policy Actor-Critic 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 |
|---|---|---|---|
| Add online meta-critic to SAC | SAC · Walker2d | Max Average Return: 6398.8 -> 7164.9 | Helped |
| Add online meta-critic to SAC | SAC · HalfCheetah | Max Average Return: 16651.8 -> 16815.9 | Helped |
| Add online meta-critic to SAC | SAC · InvPend | Max Average Return: 1000.0 -> 1000.0 | No clear effect |
| Add online meta-critic to TD3 | TD3 · TORCS | Max Average Return: 14841.7 -> 33684.2 | Helped |
| Add online meta-critic to DDPG | DDPG · HalfCheetah | Max Average Return: 8440.2 -> 15808.9 | Helped |
| Use h(π̄φ,s,a) rather than h(π̄φ) (ablation) | SAC-MC · Walker2d | Max Average Return: 7164.9 ± 151.3 -> 7423.8 ± 780.2 | Helped |
| Use h(π̄φ) rather than h(π̄φ,s,a) (ablation) | SAC-MC · Walker2d | Sum Average Return: 57,364,405 -> 61,672,039 | Helped |
| Use actor parameters directly as h input instead of actor features (ablation) | SAC-MC · Walker2d | Max Average Return: 6398.8 ± 289.2 -> 6644.3 ± 1815.6 | No clear effect |
| Replace clipped meta-loss with vanilla Lmeta (ablation) | SAC-MC · Walker2d | Max Average Return: 6398.8 ± 289.2 -> 6456.1 ± 424.8 | No clear effect |
| Increase baseline update steps to match Meta-Critic compute (ablation) | SAC · Walker2d | Max Average Return: 6398.8 -> 6363.7 | No clear effect |
| Increase baseline network parameters to match Meta-Critic (ablation) | SAC · Walker2d | Max Average Return: 6398.8 -> 6423.7 | No clear effect |
| Add online intrinsic reward learning with PPO-LIRPG | PPO · HalfCheetah | Max Average Return: 2061.5 -> 1882.6 | Hurt |
Closest papers by meaning
- Model-Based Offline Meta-Reinforcement Learning with Regularization (ICLR 2022 · similarity 0.92)
- Offline Meta-Reinforcement Learning with Advantage Weighting (ICML 2021 · similarity 0.92)
- Offline Actor-Critic Reinforcement Learning Scales to Large Models (ICML 2024 · similarity 0.91)
- Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL (ICML 2024 · similarity 0.90)
- Enhancing Online Reinforcement Learning with Meta-Learned Objective from Offline Data (AAAI 2025 · similarity 0.89)
- Diffusion Actor-Critic: Formulating Constrained Policy Iteration as Diffusion Noise Regression for Offline Reinforcement Learning (ICLR 2025 · similarity 0.88)
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.