AI Paper Atlas

Research map / Reinforcement learning / Offline RL

PerSim: Data-Efficient Offline Reinforcement Learning with Heterogeneous Agents via Personalized Simulators

Anish Agarwal, Abdullah Omar Alomar, Varkey Alumootil, Devavrat Shah, Dennis Shen, Zhi Xu, Cindy Yang

NeurIPS 2021 · Reinforcement learning › Offline RL · approach: Offline RL sample complexity theory · 2 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
Augment BCQ training data with 5 PerSim-generated trajectories per agentBCQ · CartPole (2/0.5)average reward: 43.90 -> 80.04Improved
Augment CQL training data with 5 PerSim-generated trajectories per agentCQL · CartPole (10.0/0.5)average reward: 9.300 -> 198.3Improved
Augment BCQ with PerSim trajectoriesBCQ · MountainCar 0.01average reward: -500.0 -> -68.84Improved
Augment BCQ/CQL with PerSim trajectoriesBCQ/CQL · MountainCar 0.035average reward: -500.0 -> -500.0No clear effect
Augment CQL with PerSim trajectoriesCQL · HalfCheetah (0.3/0.3)average reward: -0.600 -> -402.2Hurt
Augment BCQ with PerSim trajectoriesBCQ · HalfCheetah (0.3/1.7)average reward: -498.9 -> 571.8Improved
Use PerSim instead of PE-TS CaDM with MPCCaDM · MountainCar 0.025 rewardaverage reward: -492.3 -> -189.4Improved
Use PerSim instead of BCQ-PBCQ-P · HalfCheetah (1.7/0.3) rewardaverage reward: 2006 -> 997.0Hurt
Reduce number of training agents from N=250 to N=25 (ablation)PerSim · reward under data scarcityaverage rewardNo clear effect

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