AI Paper Atlas

Research map / Multi-agent systems and games / Cooperative multi-agent RL

Offline Multi-Agent Reinforcement Learning with Knowledge Distillation

Wei-Cheng Tseng, Tsun-Hsuan Wang, Yen-Chen Lin, Phillip Isola

NeurIPS 2022 · Multi-agent systems and games › Cooperative multi-agent RL · approach: Offline and networked MARL · 1 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
replace conventional KL distillation with relational distillation (ablation)Ours · Equal Spaceaverage return: −5.43 ± 0.51 -> −2.43 ± 0.72Improved
replace conventional KL distillation with relational distillation (ablation)Ours · Highwayaverage return: 18.33 ± 1.27 -> 23.35 ± 0.91Improved
share weights between teacher and student mapping networks (M = N) (ablation)Ours · Highwayaverage return: 23.35 ± 0.91 -> 14.92 ± 1.87Hurt
remove the mapping network (ablation)Ours · Highwayaverage return: 23.35 ± 0.91 -> 18.21 ± 1.99Hurt
remove momentum update of teacher mapping network (ablation)Ours · Highwayaverage return: 23.35 ± 0.91 -> 18.62 ± 1.98Hurt
remove KL divergence term (ablation)Ours · Highwayaverage return: 23.35 ± 0.91 -> 20.11 ± 1.88Hurt
remove KL divergence term (ablation)Ours · Grid-Worldaverage return: 2.09 ± 0.22 -> 1.99 ± 0.51No clear effect
add structural relational distillation over independent decision transformersOurs vs IDT · Highwayaverage return: 18.71 ± 1.53 -> 23.35 ± 0.91Improved
degrade offline dataset quality from good to poorOurs · Highwayaverage return: 23.35 ± 0.91 -> 17.13 ± 1.12Hurt
degrade offline dataset quality from good to poorOurs · Grid-Worldaverage return: 2.09 ± 0.22 -> 1.81 ± 0.19Hurt
use sequence modeling (MADT) instead of offline RL (MA-ICQ)baselines · Highwayaverage return: 18.01 ± 1.27 -> 18.78 ± 1.27No clear effect
online MAPPO finetuning of the pre-trained policyOurs · finetuning curvesreturnImproved

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