人工智能的10个重大数理基础问题  

Ten fundamental problems for artificial intelligence:mathematical and physical aspects

作　　者：徐宗本[1,2,3] Zongben XU(Xi’an Academy of Mathematics&Mathematical Technology,Xi’an Jiaotong University,Xi'an 710049,China;Pazhou Lab,Guangzhou 510335,China;Peng Cheng Laboratory,Shenzhen 518055,China)

机构地区：[1]西安交通大学西安数学与数学技术研究院,西安710049 [2]琶洲实验室,广州510335 [3]鹏城实验室,深圳518055

出　　处：《中国科学：信息科学》2021年第12期1967-1978,共12页Scientia Sinica(Informationis)

基　　金：国家自然科学基金重大项目(批准号:11690011,U1811461)资助。

摘　　要：本文提出并阐述人工智能研究与应用中亟待解决的10个重大数理基础问题:(1)大数据的统计学基础;(2)大数据计算的基础算法;(3)数据空间的结构与特性;(4)深度学习的数学机理;(5)非正规约束下的最优输运;(6)如何学习学习方法论;(7)如何突破机器学习的先验假设;(8)机器学习的自动化;(9)知识推理与数据学习的融合;(10)智能寻优与人工智能芯片问题.Ten fundamental to-be-solved problems for current artificial intelligence research and development are proposed and analyzed from the mathematical and physical points of view.These 10 problems include:(1)the statistical foundation problem for big data analysis,(2)the algorithm problem for basic mathematical computation in big data circumstances,(3)the understanding of structures and properties of various data spaces,(4)the mathematical theories on deep learning,(5)the investigation on the optimal transportation problem under irregular constraints,(6)the problem of how to simulate learning methodology and learn in infinite functional spaces,(7)breaking through the prior hypotheses of machine learning,(8)realizing machine learning automation from selection to creation,(9)integrating knowledge inference and data learning,(10)intelligent search and AI chips.

关 键 词：人工智能 数理基础 统计学 大数据算法 数据空间 深度学习 最优传输 学习学习方法论 机器学习假设 机器学习自动化 AI芯片 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]