摘要
数据驱动的机器学习(特别是深度学习)在自然语言处理、计算机视觉分析和语音识别等领域取得了巨大进展,是人工智能研究的热点。但是传统机器学习是通过各种优化算法拟合训练数据集上的最优模型,即在模型上的平均损失最小,而在现实生活的很多问题(如商业竞拍、资源分配等)中,人工智能算法学习的目标应该是是均衡解,即在动态情况下也有较好效果。这就需要将博弈的思想应用于大数据智能。通过蒙特卡洛树搜索和强化学习等方法,可以将博弈与人工智能相结合,寻求博弈对抗模型的均衡解。从数据拟合的最优解到博弈对抗的均衡解能让大数据智能有更广阔的应用空间。
Data-driven machine learning(especially deep learning),which is a hot topic in artificial intelligence research,has made great progress in the fields of natural language processing,computer vision analysis and speech recognition,etc.The optimization of parameters in traditional machine learning can be regarded as the process of data fitting,the optimal model on the training data set is fitted by various optimization algorithms.However,in real applications such as commodity bidding and resource allocation,the target of artificial intelligence algorithm is not an optimal solution,but an equilibrium solution,which requires the application of the game theory to big data intelligence.Combining game theory with artificial intelligence can expand the application space of big data intelligence.
作者
蒋胤傑
况琨
吴飞
JIANG Yinjie;KUANG Kun;WU Fei(College of Computer Science and Technology,Zhejiang University,Hangzhou 310027,China;Institute of Artificial Intelligence,Zhejiang University,Hangzhou 310027,China)
出处
《智能系统学报》
CSCD
北大核心
2020年第1期175-182,共8页
CAAI Transactions on Intelligent Systems
基金
国家杰出青年科学基金(61625707)
国家自然科学基金人工智能基础研究应急管理项目(61751209).
关键词
人工智能
大数据
最优拟合
神经网络结构搜索
博弈论
纳什均衡
artificial intelligence
big data
optimal fitting
neural network architecture search
game theory
Nash equilibrium
