类脑计算实际上存在两个技术层面:第1层面是"走出诺依曼框架",主要属于人工神经网络的大范畴;第2层面是"基于神经科学的计算机算法",试图超越人工神经网络框架和摆脱权值计算模型,实现对生物脑的高逼真性模拟。第2层面研究有两类方法,一类是"大科学",例如欧盟的"人类脑计划"和美国的"BRAINs"计划;另一类是"小科学",例如Numenta公司的"新皮质层模型"和Mindputer Lab的"脑的深构造网络"研究。本文总结比较了类脑计算各主要层类的方法特点,重点介绍了深构造网络的基本概念、深构造脑模型的研发进展和深构造网络技术的应用优势。
[1] 脑/生态学比较研究组. 论生命深构造:新适应主义生物学的发展及其新理论综合[J]. 前沿科学, 2009(3):55-85.
[2] Winpen H. New experimental biology:Deep structure studies II[M]. CreateSpace USA, 2012.
[3] Winpen H. The third synthesis of biology:Deep structure studies I[M]. CreateSpace, 2012.
[4] Winpen H, Haina H. The Brain's super intelligence analysis:Deep structure studies III[M]. CreateSpace, 2012.
[5] Mindputer Lab. About Mindputer project[EB/OL].[2016-02-15]. http://mindputer.org/DSD_PROGRAMS/.
[6] 华春雷. 类脑计算:打开深构造网络的大门[EB/OL].[2016-02-15]. http://blog.sciencenet.cn/home.php?mod=space&uid=2910327.
[7] Hawkins J, Blakeslee S. On intelligence[M]. Henry Holt and Company, 2004.
[8] Hawkins J. Hierarchical temporal memory (HTM) whitepaper[EB/OL].[2016-02-15]. http://numenta.com/learn/hierarchical-temporal-memorywhite-paper.html.
[9] Smith S L, Smith I T, Branco T, et al. Dendritic spikes enhance stimulus selectivity in cortical neurons in vivo[J]. Nature, 2013, 503(7474):115-120.
[10] Ellefsen K O, Mouret J B, Clune J. Neural modularity helps organisms evolve to learn new skills without forgetting old skills[J]. Plos Computational Biology, 2015, 11(4):e1004128.
[11] Mindputer Lab. New concepts of super AI[EB/OL].[2016-02-15]. http://www.mindputer.org/info/show.asp?bh=147.
[12] Mindputer Lab. Brain Computer:New conceptual design[EB/OL].[2016-02-15]. http://www.mindputer.org/info/show.asp?bh=148.
[13] Huaeren. 脑科学基础研究的四大困境:反思与评论[EB/OL].[2016-02-15]. http://www.bioon.com/biology/Class18/72513_5.shtml.
