| 中文题名： | 连续吸引子神经网络的动力学特性研究 |
| 姓名： | |
| 保密级别： | 公开 |
| 论文语种： | 中文 |
| 学科代码： | 070201 |
| 学科专业： | |
| 学生类型： | 学士 |
| 学位： | 理学学士 |
| 学位年度： | 2021 |
| 学校： | 北京师范大学 |
| 校区： | |
| 学院： | |
| 第一导师姓名： | |
| 第一导师单位： | |
| 提交日期： | 2021-05-25 |
| 答辩日期： | 2021-05-18 |
| 外文题名： | The Dynamics of continuous attractor neural network |
| 中文关键词： | |
| 外文关键词： | Population Coding ; Continuous Attractor Neural Network ; Spike Frequency Adaptation ; Anticipative Tracking |
| 中文摘要： |
阐明神经系统的信息处理机制是认识生物脑中高级功能的关键，是计算神经科学关注的重点之一。其研究结果一方面可以解释实验神经科学上的发现，同时也会对与类脑智能发展带来启示。本文总结了神经元群体编码和解码外部特征的不同方式。利用量子谐振子本征函数作为正交完备基，解析地对于连续吸引子神经网络的平滑追踪性质进行了分析。在引入神经元发放率自适应效应后，在一维和二维连续吸引子神经网络中可以产生自发的行波，当行波速度大于外界刺激速度时，将具有预测追踪效果，且在较大外界刺激运动速度的范围内，预测时间恒定。这一特性可以补偿神经系统中普遍存在的由于信息传输和处理造成的时间延迟。此外，本文还讨论了利用磁性隧道结实现含有神经元发放率自适应的连续吸引子神经网络的可行性。
﹀
|
| 外文摘要： |
Elucidating the mechanisms of information processing in the neural system is key to understanding advanced functions of the brain and developing practical information processing techniques. With the development of brain imaging techniques, experimental data enlighten computational neuroscientists to raise novel mathematical models, which can in turn shed light on experimental findings, and stimulate the development of brain-inspired intelligence. This paper summarizes the different ways of population coding for neurons to encode and decode external stimulus. The tracking capability of continuous attractor neural networks (CANNs) is derived analytically using a perturbative method. Basic eigenfunctions of one-dimensional quantum harmonic oscillator are introduced to describe different distortion modes of the Gaussian bump. With the introduction of Spike Frequency Adaptation (SFA), spontaneous traveling waves can be induced in one- and two-dimensional CANNs. When the velocity of the traveling wave is larger than that of the external stimulus, the network can achieve anticipative tracking, and the leading time is a constant over a large velocity range for the external stimulus. This property can be used to compensate the time cost for information transmission and processing that is inevitable in the neural system. In addition, the feasibility of implementing CANN with SFA by using magnetic tunnel junctions is discussed in this paper.
﹀
|
| 参考文献总数： | 46 |
| 插图总数： | 0 |
| 插表总数： | 0 |
| 馆藏号： | 本070201/21117 |
| 开放日期： | 2022-05-25 |
谷歌
