COMMA
收藏资源简介:
COMMA是由威斯康星大学麦迪逊分校和南京大学共同创建的一个多模态多智能体协作基准数据集。该数据集包含10个精心设计的协作谜题,每个谜题有数千种独特解法。数据集旨在评估多模态多智能体系统在语言交流中的协作表现,特别关注于不同智能体在信息不对称情况下的有效沟通和协作。创建过程模拟了现实世界中的协作场景,如炸弹拆除游戏,通过多轮对话和多模态信息处理来解决复杂任务。COMMA的应用领域广泛,特别是在需要多智能体协作和隐私保护的敏感数据处理场景中,如医疗保健和科学发现。
COMMA is a multimodal multi-agent collaboration benchmark dataset jointly created by the University of Wisconsin-Madison and Nanjing University. This dataset contains 10 meticulously designed collaborative puzzles, each featuring thousands of unique solutions. It is designed to evaluate the collaborative performance of multimodal multi-agent systems during linguistic communication, with a particular emphasis on effective communication and collaboration among agents in scenarios of information asymmetry. Its development process simulates real-world collaborative scenarios, such as the bomb defusal game, where agents resolve complex tasks via multi-turn dialogues and multimodal information processing. COMMA has broad application potential, especially in sensitive data processing scenarios requiring multi-agent collaboration and privacy protection, such as healthcare and scientific discovery.

- 1COMMA: A Communicative Multimodal Multi-Agent Benchmark威斯康星大学麦迪逊分校 · 2024年



