The architectonic Study of information processing system based on fabricating an active artificial brain

基于主动人工脑的信息处理系统体系结构研究

• 批准号: 14350175
• 负责人: NAKAJIMA Koji
• 金额: $ 9.54万
• 依托单位: Tohoku University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (B)
• 财政年份: 2002
• 资助国家: 日本
• 起止时间: 2002 至 2005
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-14350175/
• 关键词: Inverse function; Negative resistance; Optimization problem; Chaos; Active artificial brain; Integrated circuit; Delay; Inverse Delay

The purpose of this subject is construct a macro connecting system with silicon neuron units which have basic active properties including excitable oscillation nature to obtain the Brain like information system by using the constructive method. We proposed the Inverse function Delayed (ID) model. The model is able to increase the memory capacity for association without decrease of basin size for the memorized patterns. Furthermore, the model, which is modified into a discrete time model, achieves 100% success rate for combinatorial optimization problems with high processing speed. We have successfully carried out the TSP problems over 100 cities and etc. We also analyzed the parameter dependence of the performance of the ID model for 100% success rate of one of combinatorial optimization problems. We proposed the memory system for time sequential patterns by using the ID model. Hardware integrated system for the ID model has been investigated as an analog circuit and a digital FPGA chip. We confirmed the operation of the developed system for the ID hardware model. The ID model has been developed to the burst firing model, and its hardware system was also constructed. We are going to construct the active artificial brain system as a next research target.

本课题的目的是利用构造方法构建具有可兴奋振荡性质等基本活性特性的硅神经元单元的宏观连接系统，从而获得类脑信息系统。我们提出了反函数延迟（ID）模型。该模型能够增加关联的记忆容量，而不会减少记忆模式的盆地大小。此外，该模型修改为离散时间模型，对于组合优化问题的成功率达到100%，处理速度快。我们已经成功地解决了超过100个城市等的TSP问题。我们还分析了ID模型性能的参数依赖性，使组合优化问题之一的成功率达到100%。我们通过使用 ID 模型提出了时间顺序模式的存储系统。 ID模型的硬件集成系统已经被研究为模拟电路和数字FPGA芯片。我们确认了ID硬件模型所开发系统的运行。 ID模型已发展为点射模型，并构建了其硬件系统。我们将构建主动人工大脑系统作为下一个研究目标。

项目成果

Retrieval Property of Associative Memory Based on Inverse Function Delayed Neural Networks

基于反函数延迟神经网络的联想记忆检索特性

• 发表时间: 2005
• 期刊: IEICE Trans. Fundamentals E88-A・8
• 作者: T.Kondo 他;Yasuyuki Yamada 他;H.Li 他
• 通讯作者: H.Li 他

A New Digital Architecture of Inverse Function Delayed Neuron with the Stochastic Logic

一种新的随机逻辑反函数延迟神经元数字结构

• 发表时间: 2004
• 期刊: Proc.The 47-th IEEE International Midest Symposium on Circuits and Systems II
• 作者: M.Kinjo;et al.;T.Kondo他;Yasuyuki Yamada他;H.Li他;K.Inomata他;M.Kinjo他;S.Akimoto;A.Momoi;L.Hongge
• 通讯作者: L.Hongge

S.Akimoto: "Implementation of a new stochastic neurochip and a study on transitions between limit cycle stochastic noise"Proc. 2002 International Symposium on Nonlinear Theory and Its Applications. 917-920 (2002)

S.Akimoto：“新型随机神经芯片的实现以及极限环随机噪声之间的转换研究”Proc。

T.Haga: "Learning Algorithm for the Threshold of Nonmonotonic Neurons Composing a Boltzmann Machine"Proc. 2002 International Symposium on Nonlinear Theory and Its Applications. 687-690 (2002)

T.Haga：“构成玻尔兹曼机的非单调神经元阈值的学习算法”Proc。

Analysis of Limit-Cycles on Neural Networks with Asymmetrical Cyclic Connections Using Approximately

具有非对称循环连接的神经网络的极限环近似分析

• 发表时间: 2004
• 期刊: Proc.Of the 8^<th> KESs2004 I
• 作者: M.Kinjo;et al.;T.Kondo他;Yasuyuki Yamada他;H.Li他;K.Inomata他;M.Kinjo他;S.Akimoto;A.Momoi;L.Hongge;S.Suenaga
• 通讯作者: S.Suenaga