权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

A neural model of the binding in the human brain and its application to pattern recognition

人脑结合的神经模型及其在模式识别中的应用

基本信息

批准号：
10650353
负责人：
HARUHISA Takahashi
金额：
$ 2.18万
依托单位：
The Unicersity of Electro-Communications
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
1998
资助国家：
日本
起止时间：
1998 至 2000
项目状态：
已结题

项目摘要

The information binding in human brain is getting important to explain and to understand how human being can recognize the outer world so easily. Besides this, we expect that applying the binding mechanism to recognizer, we can get much better recognition performance. In this research we developed the neural networks that can well explain the information binding in the human brain. Based on the fact that the previous neural models are difficult to represent the spike synchronization or asynchronization, we developed two phasor neural models. One is the phase-rate oscillating neural model, in which the phase represents the sinusoidal phase. The other is covariance neural model, in which the cosine of the phase difference between two neurons represents the covariance coefficient. In both model we implemented the Hebb learning and Boltzmann learning rules to get the efficient learning. From the computer simulation results, we showed that the proposed learning rules work much more efficient than ordinal models. Applying the knowledge obtained from these models, we proposed the brain wave recognizer which shows much higher performance than previous. We also developed the theory of generalization in learning.

人脑中的信息绑定对于解释和理解人类如何如此容易地识别外部世界变得越来越重要。除此之外，我们期望将绑定机制应用到识别器上，可以获得更好的识别性能。在这项研究中，我们开发了神经网络，可以很好地解释人脑中的信息绑定。针对以往的神经网络模型难以表示峰值同步或异步的问题，我们开发了两种相量神经网络模型。一种是相率振荡神经模型，其中相位代表正弦相位。另一种是协方差神经模型，用两个神经元相位差的余弦值表示协方差系数。在这两个模型中，我们都实现了Hebb学习和Boltzmann学习规则，以获得有效的学习。计算机仿真结果表明，所提出的学习规则比常规模型更有效。利用这些模型所获得的知识，我们提出的脑电波识别器具有比以前更高的性能。我们还发展了学习中的泛化理论。