Emergence of machine language: towards symbolic intelligence with neural networks  

作　　者：Yuqi Wang Xu-Yao Zhang Cheng-Lin Liu Tieniu Tan Zhaoxiang Zhang 

机构地区：[1]State Key Laboratory of Multimodal Artificial Intelligence Systems,Institute of Automation,Chinese Academy of Sciences,China

出　　处：《National Science Review》2024年第4期21-24,共4页国家科学评论（英文版）

基　　金：This work is supported in part by the 2035 Innovation Program of CAS,the National Key R&D Program of China(2022ZD0160102);the National Natural Science Foundation of China(61836014,U21B2042,62072457 and 62006231).

摘　　要：Representation learning is a core issue in artificial intelligence(AI).Currently,there exists a disparity in the choice of representation between humans and machines.Humans rely on discrete language for communication and learning,whereas machines utilize continuous features for computation and representation.Discrete symbols are low-dimensional,decoupled and offer robust reasoning abilities,while continuous features are high-dimensional,coupled and possess remarkable abstracting capabilities.In recent years,deep learning[1]has developed the idea of continuous representation to the extreme,using billions of parameters to achieve high accuracies.Although this is reasonable from a statistical perspective,it has other major problems,such as a lack of interpretability,poor generalization and being easily attacked.Both paradigms have strengths and weaknesses,and a better choice is to seek reconciliation.

关 键 词：SYMBOLS utilize SYMBOLIC 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]