Asynchronous Pulse-type Chaotic Neural Networks with Hardware Models

大脑各子系统具有高度的信息处理能力，即识别和学习。然而，信息处理的功能还没有明确。因此，为了阐明生物神经网络的信息处理功能，并将其应用于工程问题，人们对各种神经元模型和人工神经网络进行了研究。利用VLSI技术构建脑型信息处理系统需要人工神经网络的支持。在本研究中，我们讨论了异步脉冲型混沌神经网络的硬件模型。结果表明：(1)在我们提出的硬件活动树突模型中，活动树枝模型具有相似的生物反向传播特性(参考文献1)。(2)我们提出了用A型负阻器件实现多值记忆细胞的方法(参考文献2)。(3)我们构造了一个短期记忆电路，并使用硬件环神经网络(参考文献3)验证了时间模式识别电路的记忆模式。(4)研究了STDP对提取隐藏在波动中的相位信息能力的影响。我们重点研究了脉冲时序相关突触可塑性(STDP)，并利用STDP构建了一个脉冲型硬件神经元模型的神经网络。我们发现，隐藏在波动中的相位信息是通过STDP的效应来揭示的，这使得解码突触权重成为可能。此外，我们证明了利用带有STDP的神经网络提取隐藏在波动中的相位差，代表突触权重的强化部分。(5)证明了通过给CPG(中央模式生成器)模型一个脉冲的外部输入来产生和转换振荡模式是可能的(参考文献4)。

Brain subsystems have a high degree of information processing ability, namely recognition and learning. However, the information processing functions have not been clarified as yet. As a result, various neuron models and artificial neural networks have been studied in order to clarify the information processing functions of biological neural networks, and apply them to engineering problems. Artificial neural networks performing similarly to the human brain are required for constructing an information processing system of brain-type using the VLSI technology.In this study, we discuss asynchronous pulse-type chaotic neural networks with hardware models.Results,(1)In our proposed hardware active dendrite model, it is shown clearly that the active dendrite model has similar biological backpropagation characteristics (References No.1).(2)We propose the CMOS implementation of a multiple valued memory cell using A-shaped negative resistance devices for plastic synapses (References No.2).(3)We construct a short-term memory circuit, and we verify the memory patterns of the temporal pattern recognition circuit using hardware ring neural networks (References No.3).(4)We investigate the effect of STDP on the ability to extract phase information buried in fluctuation. We focus on spike timing dependent synaptic plasticity (STDP), and we construct neural networks from a pulse-type hardware neuron model using STDP. We show that phase information buried in fluctuation is revealed by the effect of STDP, making it possible to decode the synaptic weight. Moreover, we show that it is possible to extract the phase difference buried in fluctuation representing the reinforcement part of the synaptic weight, using neural networks with STDP.(5)It is shown that generation and transition of oscillation patterns are possible by giving external inputs of one pulse to the CPG (Central Pattern Generator) model (References No.4).