ART2wNF及其稳定性——可塑性动态平衡性能分析

陈众; 莫红

doi:10.3724/SP.J.1004.2013.01381

ART2wNF及其稳定性——可塑性动态平衡性能分析

doi: 10.3724/SP.J.1004.2013.01381

陈众,
莫红

1.
长沙理工大学电气与信息工程学院长沙 410004

基金项目:

国家自然科学基金(61074903, 61233088);中国科学院自动化研究所复杂系统管理与控制国家重点实验室开放课题;长沙理工大学青年英才计划资助

详细信息

作者简介:
莫红长沙理工大学电气与信息工程学院副教授. 2004 年获得中国科学院研究生院工学博士学位. 主要研究方向为语言动力系统与智能计算. E-mail: mohong72@gmail.com

计量
- 文章访问数: 1395
- HTML全文浏览量: 24
- PDF下载量: 1278
- 被引次数: 0
出版历程
- 收稿日期: 2012-06-27
- 修回日期: 2013-03-04
- 刊出日期: 2013-08-20

ART2wNF and Its Stability-plasticity Dynamic Trade-off Analysis

CHEN Zhong,
MO Hong

1.
School of Electric and Information Engineering, Changsha University of Science and Technology, Changsha 410004

Funds:

Supported by National Natural Science Foundation of China (61074903, 61233088), Open Project of State Key Laboratory of Management and Control for Complex Systems, Institute of Automation, Chinese Academy of Science, and Youth Talent Support Plan of Changsha University of Science and Technology

摘要

摘要: 稳定性-可塑性两难问题的核心是系统如何在不削弱或忘记已学习模式的同时, 自适应地学习新事物. 目前公认自适应谐振理论(Adaptive resonance theory, ART) 能够部分解决稳定性-可塑性两难问题, 但依然存在学习受样本输入顺序影响大, 且存在学习中心渐变样本时, 带来的所谓模式漂移的问题. 受进化生物学关于人类学习的幼态延续特征的启发, 本文为每个F2层节点配备活跃度指示器λ, 并将其反馈回F1层参与STM(Short term memory) 向量的计算, 使这种新型ART2 网络在行为特征上具备幼态延续的显著特征, 本文称之为ART2wNF (Adaptive resonance theory with neoteny feature).论文从理论上证明算法的可行性, 并通过分析对随机生成样本集合的学习过程, 对比了ART2wNF 算法与常规ART2 网络在可塑性、稳定性方面的差异以及ART2wNF 在克服样本输入顺序影响等方面的优势.
- 稳定性-可塑性两难问题 /
- 幼态延续 /
- 自适应谐振理论 /
- ART2网络 /
- ART2wNF
Abstract: Stability-plasticity dilemma is how to build a system that is adaptive enough to learn new things while not diluting/forgetting previously learned patterns. It is well known that ART (adaptive resonance theory) network can partially solve the stability-plasticity dilemma, but the behavior of ART network is uncertain due to the input order of samples and the pattern drift problem which is also notable for patterns with gradually changed center. Inspired by the human neoteny phenomenon discussed in evolution biology, in order to record the stimulating degree, we suggest that each node in F2 layer be accompanied by an activity indicator λ, which is also fed back to F1 layer as a parameter of the calculation of STM (short term memory) vectors. The modified ART2 network has the remarkable feature of neoteny during learning process and is called ART2wNF (adaptive resonance theory with neoteny feature) in this paper. The feasibility of the arithmetic is theoretically proved for the introduction of λ at first. Then the performance and distinctness of ART2wNF in stability and plasticity are compared with ART2 by analyzing the learning process for randomly generated samples. It also shows the distinctive ability to overcome the shortage of ART2 caused by different input orders.
- Stability-plasticity dilemma /
- neoteny /
- adaptive resonance theory (ART) /
- ART2 network /
- ART2wNF

HTML全文

参考文献(24)

[1]	Zhou Zhi-Hua, Chen Zhao-Qian, Chen Shi-Fu. Review of adaptive resonance theory. Computer Science, 1999, 26(4): 54-56, 79 (周志华, 陈兆乾, 陈世福. 自适应谐振理论综述. 计算机科学, 1999, 26(4): 54-56, 79)
[2]	Carpenter G A, Grossberg S. A massively parallel architecture for a self-organizing neural pattern recognition machine. Computer Vision, Graphics, and Image Processing, 1987, 37(1): 54-115
[3]	Carpenter G A, Grossberg S. ART 2: self-organization of stable category recognition codes for analog input patterns. Applied Optics, 1987, 26(23): 4919-4930
[4]	Carpenter G A, Grossberg S, Rosen D B. ART2-A: an adaptive resonance algorithm for rapid category learning and recognition. Neural Networks, 1991, 4(4): 493-504
[5]	Carpenter G A, Grossberg S. ART 3: hierarchical search using chemical transmitters in self-organizing pattern recognition architectures. Neural Networks, 1990, 3(2): 129-152
[6]	Carpenter G A, Grossberg S, Reynolds J H. ARTMAP: supervised real-time learning and classification of nonstationary data by a self-organizing neural network. Neural Networks, 1991, 4(5): 565-588
[7]	Yao Ming-Hai, Qu Xin-Yu, Li Jia-He, Gu Qin-Long, Tang Li-Ping. Study on Q-learning algorithm based on ART2. Control and Decision, 2011, 26(2): 227-232 (姚明海, 瞿心昱, 李佳鹤, 顾勤龙, 汤丽平. 基于ART2的Q学习算法研究. 控制与决策, 2011, 26(2): 227-232)
[8]	Chen Zhong, Li Wei. Moving object detection based on ART2 network. Journal of Changsha University of Science and Technology (Natural Science), 2012, 9(1): 54-59 (陈众, 李维. 基于ART2网络的运动目标检测. 长沙理工大学学报(自然科学版), 2012, 9(1): 54-59)
[9]	He Tao, Zhao Ying-Ying. An improved character recognition algorithm based on ART2 neural network. Journal of Guilin University of Electronic Technology, 2012, 32(3): 237-239 (何滔, 赵莹莹. 一种改进的基于ART2神经网络的文字识别算法. 桂林电子科技大学学报, 2012, 32(3): 237-239)
[10]	Noh M S, Hong D S. Implementation of remote monitoring system for prediction of tool wear and failure using ART2. Journal of Central South University of Technology, 2011, 18(1): 177-183
[11]	Yang Xu-Hua. Study on Neural Networks Machine and Its Application in Control[Ph.D. dissertation], Zhejiang University, China, 2004 (杨旭华. 神经网络及其在控制中的应用研究[博士学位论文], 浙江大学, 中国, 2004)
[12]	Brüne M. Neoteny, psychiatric disorders and the social brain: hypotheses on heterochrony and the modularity of the mind. Anthropology & Medicine, 2000, 7(3): 301-318
[13]	Chen Zhong, Cai Zi-Xing, Ye Qing. Color pixel categorization based on ART2 network. Journal of Image and Graphics, 2008, 13(4): 634-641 (陈众, 蔡自兴, 叶青. 基于ART2网络的彩色像素分析及其应用. 中国图象图形学报, 2008, 13(4): 634-641)
[14]	Chen Zhong. Realization of Cognitive Developmental Process in Schema Theory with ART2 Network and Its Application[Postdoctoral Report], Central South University, China, 2007 (陈众. 图式理论关于认知发展过程描述的ART2网络实现及其应用. 中南大学控制科学与工程博士后流动站出站报告, 中国, 2007)
[15]	Chen Z, Liu R, He L. Supervised ART2 network with the features of the developmental process of human cognition. In: Proceedings of the 7th World Congress on Intelligent Control and Automation. Chongqing, China: IEEE, 2008. 3754-3759
[16]	Zuo Ren-Xia, Li Wei-Qi. Pi-Ya-Jie. Fa Sheng Ren Shi Lun Wen Xuan. Shang hai: East China Normal University Press, 1991(左任侠, 李其维. 皮亚杰发生认识论文选. 上海: 华东师范大学出版社, 1991)
[17]	Chen Z, Wang M, He L. Neoteny learning rule of ART2 network and its function. In: Proceedings of the 6th International Conference on Natural Computation. Yantai, China: IEEE, 2010. 407-411
[18]	Xue G, Dong Q, Chen C S, Lu Z L, Mumford J A, Poldrack R A. Greater neural pattern similarity across repetitions is associated with better memory. Science, 2010, 330(6000): 97-101
[19]	Tenenbaum J B, Kemp C, Griffiths T L, Goodman N D. How to grow a mind: statistics, structure, and abstraction. Science, 2011, 331(6022): 1279-1285
[20]	Gopnik A, Glymour C, Sobel DM, Schulz LE, Kushnir T, Danks D. A theory of causal learning in children: causal maps and Bayes nets. Psychological Review, 2004, 111(1): 3-32
[21]	Kemp C, Tenenbaum J B. Structured statistical models of inductive reasoning. Psychological Review, 2009, 116(1): 20-58
[22]	Wang Fei-Yue. Fundamental issues in research of computing with words and linguistic dynamic systems. Acta Automatica Sinica, 2005, 31(6): 844-852(王飞跃. 词计算和语言动力学系统的基本问题和研究. 自动化学报, 2005, 31(6): 844-852)
[23]	Mo Hong, Wang Fei-Yue, Xiao Zhi-Quan, Chen Qian. Stabilities of linguistic dynamic systems based on interval type-2 fuzzy sets. Acta Automatica Sinica, 2011, 37(8): 1018-1023(莫红, 王飞跃, 肖志权, 陈茜. 基于区间二型模糊集合的语言动力系统稳定性. 自动化学报, 2011, 37(8): 1018-1023)
[24]	Mo Hong. Linguistic dynamic orbits in the time varying universe of discourse. Acta Automatica Sinica, 2012, 38(10): 1585-1594(莫红. 时变论域下的语言动力学轨迹. 自动化学报, 2012, 38(10): 1585-1594)