摘要
当前,随着博弈环境复杂多变而引入深度学习模型如深度卷积神经网络,以辅助提升人员对博弈态势的认知和决策水平。然而,将深度学习引入博弈态势理解的同时,也引入了人工智能中的数据不确定度和认知不确定度,导致人工智能预测和决策结果存在发散性等问题。基于测量不确定度评定方法对博弈态势理解中的测量过程进行不确定度关键要素分解、提取及测量模型构建。试验结果表明,基于GUM的物理测量方法能有效对博弈态势的认知不确定度进行准确高效的测量和评估。最后,基于蒙特卡罗方法对提出的博弈态势认知不确定度新质测量方法进行验证,表明提出方法的准确性和适用性。
At present,with the complex and changeable game environment,deep learning models such as deep convolutional neural networks are introduced to assist in improving personnel's cognition and decision-making level of the game situation.However,when deep learning is introduced into game situation understanding,it also introduces data uncertainty and cognitive uncertainty in artificial intelligence,which leads to problems such as divergence of artificial intelligence prediction results.Key elements of uncertainty in the measurement process of game situation understanding are decomposed,extracted and measurement modeling constructed based on the measurement uncertainty evaluation method.The experimental results show that the physical measurement method based on GUM can effectively measure and evaluate the cognitive uncertainty of game situation accurately and efficiently.Finally,based on Monte Carlo method,the proposed new qualitative measurement method of game situation cognition uncertainty is verified,which shows the accuracy and applicability of the proposed method.
作者
王永光
孙静
张楠
张修建
WANG Yongguang;SUN Jing;ZHANG Nan;ZHANG Xiujian(Beijing Aerospace Institute for Metrology and Measurement Technology,Beijing,100076;Key Laboratory of Artificial Intelligence Measurement and Standards for State Market Regulation,Beijing,100076)
基金
国家重点研发计划(2022YFF0605200)。
关键词
深度学习
态势理解
认知不确定度
测量不确定度
蒙特卡罗方法
deep learning
situation understanding
cognitive uncertainty
measurement uncertainty
Monte Carlo method
作者简介
王永光(1987-),男,博士,工程师,主要研究方向为人工智能计量与测试技术、深度学习不确定性估计技术、人工智能安全可信测评技术等。;孙静(1993-),女,工程师,主要研究方向为人工智能计量与测试技术、智能特性与安全可信测评技术、人工智能测评体系建设与标准规范等。;张楠(1990-),女,工程师,主要研究方向为软件工程、人工智能与机器人、人工智能计量与测试技术等。;张修建(1984-),男,研究员,主要研究方向为人工智能计量与测试技术、对抗安全性测评技术、人工智能测评规范等。
