AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

On Effective Scheduling of Model-based Reinforcement Learning

Hang Lai, Jian Shen, Weinan Zhang, Yimin Huang, Xing Zhang, Ruiming Tang, Yong Yu, Zhenguo Li

NeurIPS 2021 · Reinforcement learning › Policy gradient and actor-critic · approach: Model-based RL with guarantees · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (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 changedWhereMeasuredEffect
Use AutoMBPO to jointly schedule the real-data ratio, model-training frequency, policy-training iteration, and rollout lengthMBPO · Hopper, Ant, Humanoid, HopperBullet, Walker2dBullet, HalfcheetahBulletaverage returnImproved
Use AutoMBPO scheduling instead of PBT or RoRMBPO · Continuous control environmentsaverage returnImproved
Increase the real-data ratio over trainingMBPO · Six continuous control environmentsaverage returnImproved
Schedule only the real-data ratio while keeping other hyperparameters at the original MBPO settings (ablation)MBPO · Almost all evaluated tasksaverage returnImproved
Schedule only policy-training iterations (ablation)MBPO · Hopperaverage returnImproved
Schedule only policy-training iterations (ablation)MBPO · Antaverage returnHurt
Schedule only model-training frequency (ablation)MBPO · Antaverage returnImproved
Schedule only model-training frequency (ablation)MBPO · Hopperaverage returnHurt
Use SAC with policy training iteration matched to MBPO (SAC(20)) instead of SAC(1) (ablation)SAC · Ant and Walker2dBulletaverage returnHurt
Increase the initial real-data ratio or policy-training iteration for AutoMBPO (ablation)AutoMBPO · Three PyBullet environmentsaverage returnNo clear effect
Use a larger initial policy-training iteration and wider iteration limit (ablation)AutoMBPO · Three PyBullet environmentsaverage returnNo clear effect
Transfer a hyper-controller trained on another PyBullet task without fine-tuning (ablation)AutoMBPO · PyBullet environmentsaverage returnImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy