The impact of behavioral diversity in multi-agent reinforcement learning
원문 초록
Many of the world's most pressing issues, such as climate change and global peace, require complex collective problem-solving skills. Recent studies indicate that diversity in individuals' behaviors is key to developing such skills and increasing collective performance. Yet behavioral diversity in collective artificial learning is understudied, with today's machine learning paradigms commonly favoring homogeneous agent strategies over heterogeneous ones, mainly due to computational considerations. In this work, we employ diversity measurement and control paradigms to study the impact of behavioral heterogeneity in several facets of multi-agent reinforcement learning. Through experiments in team play and other cooperative tasks, we show the emergence of unbiased behavioral roles that improve team outcomes; how behavioral diversity synergizes with morphological diversity; how diverse agents are more effective at finding cooperative solutions in sparse reward settings; and how behaviorally heterogeneous teams learn and retain latent skills to overcome repeated disruptions. Overall, our results indicate that, by controlling diversity, we can obtain non-trivial benefits over homogeneous training paradigms, demonstrating that diversity is a fundamental component of collective artificial learning, an insight thus far overlooked.
함께 살펴볼 논문
공통 태그를 근거로 연결했습니다. 실제 인용 관계나 논문의 우수성을 뜻하지 않습니다.
When Is Diversity Rewarded in Cooperative Multi-Agent Learning?
One Step is Enough: Multi-Agent Reinforcement Learning based on One-Step Policy Optimization for Order Dispatch on Ride-Sharing Platforms
Step-wise Policy for Rare-tool Knowledge (SPaRK): Offline RL that Drives Diverse Tool Use in LLMs
Exploring Design of Multi-Agent LLM Dialogues for Research Ideation
MS-DPPs: Multi-Source Determinantal Point Processes for Contextual Diversity Refinement of Composite Attributes in Text to Image Retrieval
Prompt-Free Conditional Diffusion for Multi-object Image Augmentation
토론
이 논문과 연결된 포럼 글이 아직 없습니다.