数据驱动与知识引导结合下人工智能算法模型  

Artificial Intelligence Algorithms Based on Data-driven and Knowledge-guided Models

作　　者：金哲 张引[1] 吴飞[1] 朱文武 潘云鹤[1] JIN Zhe;ZHANG Yin;WU Fei;ZHU Wenwu;PAN Yunhe(College of Computer Science and Technology,Zhejiang University,Hangzhou 310012,China;Department of Computer Science and Technology,Tsinghua University,Beijing 100084,China)

机构地区：[1]浙江大学计算机科学与技术学院,杭州310012 [2]清华大学计算机科学与技术系,北京100084

出　　处：《电子与信息学报》2023年第7期2580-2594,共15页Journal of Electronics & Information Technology

基　　金：中国工程科技知识中心项目(CKCEST-2021-1-8);国家自然科学基金(62037001)。

摘　　要：当前人工智能的学习模式主要以数据驱动为主要手段,以深度神经网络为主流的机器学习算法取得了显著进展。但是这种数据驱动的人工智能手段依然面临数据获取成本高、可解释性弱、鲁棒性不强等不足。该文认为在现有机器学习算法中引入先验假设、逻辑规则和方程公式等知识,建立数据和知识双轮驱动的人工智能方法,将推动更通用计算范式的变革创新。该文将可用于引导人工智能算法模型知识归纳为4种——逻辑知识、视觉知识、物理定律知识和因果知识,探讨将这些知识与现有数据驱动模型相互结合的典型方法。Nowadays,artificial intelligence is in the era of big data-driven manner.Machine learning algorithms with deep neural networks as the mainstream have achieved great development and achievements.However,data-driven artificial intelligence still faces problems such as the cost of annotating data,the lack of interpretability,and the weak robustness.The Introduction of knowledge such as prior hypothesis,logic rules and physical equations into existing machine learning algorithms will build artificial intelligence approaches powered by both data and knowledge which could promote innovations of computing paradigm.Four types of knowledge(logical knowledge,visual knowledge,laws of physics knowledge and causal knowledge)that can be used to guide artificial intelligence algorithm models are summarized in thus paper,and typical approaches to guide the combination of these knowledge with data-driven models are discussed.

关 键 词：人工智能 大数据 数据驱动 知识引导 

分 类 号：TN911[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]