Can I read Learning and Planning Multi-Agent Tasks Via An MoE-based World Model ZijieZhao2025NIPS on EtoBox?

Learning and Planning Multi-Agent Tasks Via An MoE-based World Model ZijieZhao2025NIPS by f2work is a document available to read on EtoBox.

What is Learning and Planning Multi-Agent Tasks Via An MoE-based World Model ZijieZhao2025NIPS about?

The document presents M3W, a novel approach to multi-task multi-agent reinforcement learning (MT-MARL) that utilizes a mixture-of-experts (MoE) based world model for learning and planning. It addresses the challenge of knowledge reuse across similar tasks while preventing gradient conflicts from dissimilar tasks by modeling shared dynamics instead of separate policies. M3W demonstrates superior performance and sample efficiency on benchmark tasks, establishing itself as the first MoE-based multi-task world

Author

f2work

Language

EN