AIGC军事大模型评估体系框架研究  

Research on the framework of AIGC military large model evaluation system

作　　者：张龙 王数 雷震 冯轩铭 杨波 Zhang Long;Wang Shu;Lei Zhen;Feng Xuanming;Yang Bo(System Engineering Institute of the Academy of Military Sciences,Beijing 100101,China)

机构地区：[1]军事科学院系统工程研究院,北京100101

出　　处：《战术导弹技术》2025年第1期42-52,共11页Tactical Missile Technology

摘　　要：生成式人工智能(AI-Generated Content,AIGC)关键技术突破推动多模态大语言模型(Multimodal Large Language Models,MLLMs)军事垂直领域应用过程中存在评估体系评估指标不够健全的问题,为解决此问题,采用自顶向下正向设计与自底向上聚合评估相结合的方法,构建包含智能化军事需求—智能化场景任务—系统性能评估—体系效能评估的“四域”,与基础支撑服务—算法指标体系—综合安全防护的“三维”军事大模型评估体系框架,提出评估大模型的主要维度、关键指标和基本流程,并定性定量相结合给出相应评估指标体系,为军事大模型赋能装备体系和作战效能提供评估支撑。To solve the insufficiently perfect evaluation framework and comprehensive assessment metrics for the application of Multimodal Large Language Models(MLLMs) in military vertical domains, which makes significant breakthroughs in key technologies for AI-Generated Content(AIGC), a combined top-down deconstruction and bottom-up aggregation evaluation method is adopted. An overall evaluation framework for military large models is constructed, which includes four domains: intelligent military requirements, intelligent scenario tasks, system-level performance evaluation, and system-level effectiveness evaluation. Additionally,three dimensions of foundational support services, algorithmic metrics systems, and integrated security protection are encompassed in the evaluation framework. Major perspectives, critical indicators, and basic procedures for evaluating large models and the corresponding evaluation combining qualitative and quantitative are proposing, provided an evaluation foundation for the enhancement of equipment systems and combat effectiveness through military large models.

关 键 词：生成式人工智能(AIGC) 多模态大语言模型(MLLMs) 军事大模型 智能化 评估 体系效能 体系框架 

分 类 号：E917[军事]