Computational Study on Recognition and Memory in the Asymmetric <symmetric Neural Networks

非对称<对称神经网络识别与记忆的计算研究

• 批准号: 12680379
• 负责人: ISHII Naohiro
• 金额: $ 2.24万
• 依托单位: Nagoya Institute of Technology
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 2000
• 资助国家: 日本
• 起止时间: 2000 至 2001
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-12680379/
• 关键词: Neural Network; Asymmetric network; Nonlinear Analysis; Construction of two asymmetric network; Moving stimulus; Retinal networks; Higher-order correlation; Optimal pathway; 生物ニューラルネット; 非対称ニューラルネット; 空間相関; センサー情報の統合; 高速学習; 追加学習

Biological neural networks are different from artificial neural networks. There are functionally different neurons in the networks. In the field of neuro-physiology , neurons are classified into cells with several types of function by the experiments. We classify the biological neural network morphologically into two types of sub-networks. One is the symmetrical network, while the other is the asymmetrical network. We extended this asymmetrical network to the generalized asymmetrical network to clarify the characteristics of the asymmetrical networks. The generalized asymmetrical network extended to the parallel network with the odd nonlinear pathway of neurons and the even nonlinear pathway of neurons. The function of this generalized asymmetrical networks, is analyzed by Wiener analysis method. Then, the temporal correlations are important in the asymmetrical network. We derived this generalized network is reduced to the asymmetrical network, which consists of the parallel network with the linear pathway, which shows an odd nonlinearity, and with the nonlinear pathway, which shows the 2nd order nonlinearity. We clarified that the biological network is composed with two asymmetrical sub-networks, which realizes the spacial and temporal correlation as the stimulus information transmitted to the central neural network.

生物神经网络不同于人工神经网络。网络中有功能不同的神经元。在神经生理学领域，通过实验将神经元分为具有多种功能的细胞。我们将生物神经网络从形态学上分为两类子网络。一种是对称网络，另一种是非对称网络。我们将这种非对称网络推广到广义非对称网络，以阐明非对称网络的特性。将广义非对称网络扩展到具有神经元奇非线性路径和神经元偶非线性路径的并行网络。用维纳分析法分析了这种广义非对称网络的功能。因此，在非对称网络中，时间相关性是重要的。我们推导出这种广义网络被简化为非对称网络，它由具有奇非线性的线性通路和具有二阶非线性的非线性通路的并联网络组成。阐明了生物网络是由两个不对称的子网络组成的，在刺激信息传递到中枢神经网络的同时，子网络实现了空间和时间的关联。

项目成果

N.Ishii, H.Nakamura,M.Ohta: "Correlation Functions in Nonlinear Neural Networks"Proc. Of the Seventh International Artificial Life & Robotics. 148-151 (2002)

N.Ishii、H.Nakamura、M.Ohta：“非线性神经网络中的相关函数”Proc。

N.Yamaguchi, Yamauchi, N.Ishii: "An Incremental Learning Method using Weighted Magnitude"Proc.2000 IEEE Conf.on Ind.Electronics C & I. (IECON). 1189-1194 (2000)

N.Yamaguchi、Yamauchi、N.Ishii：“使用加权幅度的增量学习方法”Proc.2000 IEEE Con​​f.on Ind.Electronics C

H.Takeuchi, K.Yamauchi, N.Ishii: "Self-supervisees Leaning and Recognition by Integrating Int."Proc. IEEE Conf. on Ind. Electronics C&I. (IECON). 1195-1200 (2000)

H.Takeuchi、K.Yamauchi、N.Ishii：“通过整合智力来自我监督学习和识别。”Proc。

N.Yamaguchi, Yamauchi, N.Ishii: "An Incremental Learning Method using Weighteal Magnitude"Proc. IEEE Conf. on Ind. Electronics C&I. (IECON). 1189-1194 (2000)

N.Yamaguchi、Yamauchi、N.Ishii：“使用权重幅度的增量学习方法”Proc。

M.Ohta, N.Ishii, M.Nakamura: "Optimization with Linear Constraints In the Neural Networks"Lecture Notes in Computer Science(Springer Verlag). Vol2083. 561-569 (2001)

M.Ohta、N.Ishii、M.Nakamura：“神经网络中线性约束的优化”计算机科学讲义（Springer Verlag）。