|Place of Conferral||北京|
|Keyword||常识知识处理 类脑脉冲神经网络 多任务|
随着对大脑结构的深入研究，每个脑区的功能都有部分定性的结论。而在神经科学家的努力之下，构成各个脑区的生物神经元和生物神经元之间的突触活动模式，也有了初步的数学表达。鉴于脑中神经元不仅数量较为庞大，而且种类相当繁多，所以一直以来生物神经元如何在各个脑区中组织并发挥相应的作用都是一个未解的难题。它的解决将促进对Human Level AI的研究，同时也会对通用型的人工智能系统的实现有所帮助。而且在解决该难题中产生的结论也能和生物神经系统的研究结论相互启发和相互印证。
常识知识处理被人工智能科学家认为是实现Human Level AI的关键，是通用人工智能系统的所应具备的基本认知能力。常识知识是个人对于社会的运行规则和自然环境的变化规律的认识，是个人知识体系的基础，对个人做出的选择有潜在的影响。通过计算建模的方法提出人脑如何利用最基本的神经元组件和突触学习法则以及脉冲神经网络处理常识知识，对于构建类脑的语义理解系统，使机器真正理解人类语言具有较高的参考意义。
|Other Abstract||In recent years, functions of each brain region have been identified to some extent, with advancement of research in the structure of human brain. In the meanwhile, the sustained efforts of neuroscientists have led to basic mathematical formulas, describing biological neurons that compose the brain, and behavior patterns of synapses between biological neurons. Due to the large quantity and many different kinds of biological neurons, it is always a tough problem to understand how biological neurons organize and work in each brain region. The solution of this problem will promote the study of human-level AI and make a contribution to the realization of Artificial General Intelligence. Conclusions made during the period of solving this problem may inspire and mutually confirm with those conclusions from research on biological neural system.|
Commonsense Knowledge processing is considered as a key to achieve human level AI by artificial intelligence researchers, and is a basic cognitive ability that the artificial general intelligence system is endowed with. Commonsense knowledge is the knowledge of individual on operation rules of our society and evolution laws of nature. And it is the basis of individual knowledge hierarchy, having underlying influence on their decisions making in every day. Computational modeling will help us to understand how to use the most basic elements of biological neurons, rules of synaptic learning and spiking neural network to deal with common sense knowledge. It also has a significance of reference in constructing a brain-inspired semantic understanding system and making robots really understand human language.
In this dissertation, we provide two types of spiking neural networks, i.e. synapse-growth and synapse-inhibition spiking neural networks, and use them respectively to process three kinds of tasks of commonsense knowledge, including knowledge induction, reasoning through transitive relation, and generating conceptual commonsense knowledge through entity conceptualization. The experimental results validates the proposed models. In addition, we conducted the corresponding experiments on ConceptNet and Baidu Baike knowledge base, and obtained their performance in processing large amounts of data. The potential problems of our proposal and future research are also given in this dissertation.
The innovation of this dissertation is that we propose mathematical definitions of three kinds of tasks of commonsense knowledge and design two kinds of spiking neural network to process those tasks. The ability of handling three different types of tasks simultaneously indicates that these spiking neural networks can be the core component of brain-inspired spiking neural networks, which are used to process common sense knowledge.
|First Author Affilication||Institute of Automation, Chinese Academy of Sciences|
|唐剑波. 面向常识知识处理的类脑脉冲神经网络计算模型[D]. 北京. 中国科学院研究生院,2017.|
|Files in This Item:|
|面向常识知识处理的类脑脉冲神经网络计算模（1789KB）||学位论文||暂不开放||CC BY-NC-SA||Application Full Text|
|Recommend this item|
|Export to Endnote|
|Similar articles in Google Scholar|
|Similar articles in Baidu academic|
|Similar articles in Bing Scholar|
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.