超大规模可解释机器智能系统建设

Construction of Super Large Interpretable Machine Intelligence System

为了在现实的计算机系统软硬件能力基础上,构建超越人类认知和智慧水平的超级智能系统,提出了超大规模可解释智能系统工程化建设方案。首先,明确了模式检索为核心的基本实现思路,并以数学应用题类人自动求解为例,在总体运行基本框架的基础上,突出了基于语义继承与重载的全局语义分析、自主学习和样本数据近似计算等创新点;阐述了建设标准规范体系、渐进学习进化、有限语义和逻辑延展、大规模模式匹配和运行监测数据体系等基础理论问题。其次,讨论了以技术路线、全局数据池、语义距离计算与特征类比方法、功能处理节点、特征判断节点以及负载均衡并行计算模型等工程化实施的关键技术。最后,介绍了下一步深化研究的方向。

For building the super-smart system that could surpass human cognition and intelligence ability based on the computer software and hardware capabilities this day,we proposed the engineering construction plan for super large interpretable intelligent systems.Firstly,we clarified the core implementation approach with pattern retrieval.Taking humanoid automatic resolving of mathematic application problem as the example,with the running architecture,innovations such as global semantic analysis based on semantic inheritance and overloading,autonomous learning,and approximate calculation for sample data were highlighted;Explained the basic theoretical issues,such as the standard specification preparing,progressive evolving,limited semantics and logic extending,large-scale pattern matching,and running monitoring.Secondly,key technologies for engineering implementation were discussed,including technology roadmap,global data pool,semantic distance calculation and feature analogy methods,function processing nodes,feature judgment nodes,and workload balance parallel computing models.Finally,the directions for further research were introduced.