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

A Biologically Plausible Spiking Neuron in Hardware

硬件中生物学上合理的尖峰神经元

基本信息

批准号：
EP/F05551X/1
负责人：
Steve Hall
金额：
$ 55.12万
依托单位：
University of Liverpool
依托单位国家：
英国
项目类别：
Research Grant
财政年份：
2008
资助国家：
英国
起止时间：
2008 至无数据
项目状态：
已结题

来源：
https://gtr.ukri.org/projects?ref=EP%2FF05551X%2F1
关键词：
Biologically Plausible Spiking Neuron Hardware

项目摘要

Neuroscientists now know that the human brain is made up of millions of small units, called neurons, which are connected to each other in a very complex way. These neurons carry out relatively simple calculations using the information that enters the brain from our senses (eyes, touch, etc,) and the result of these calculations is passed on to other neurons as small electrical signals. Because each neuron performs simple calculations, it is believed that very complex calculations, such as recognizing someone, can be achieved when millions of neurons are connected together to form a network, as is the case in the human brain. Engineers and scientists are interested in how the brain carries outthese calculations because the computing power of the brain far exceeds that of any man made machine, such as the desktop computer. Much of the processing the brain is learned over time. Therefore, to understand how the brain learns to perform complex calculations, engineers and scientists are continually trying to build models of the brain, called artificial neural networks. Much of this modeling is carried out using computers or electronic circuits that mimic neuron behavior. The problems facing engineers and scientists in designing electronic neurons are: 1) designing circuits that behave like neurons and 2) making the circuits small enough so that millions of them can be placed on a silicon chip and 3) these neurons must consume minimal power. Since there are no available electronic component that can mimic the components of a neurons, what is required is the development of a new electronic components with small physical dimensions that operates just like real neurons and consume miminal power. This is what we are trying to achieve. The project aims to develop an electronic neuron that has the capability of mimicking a biological neuron but yet consume minimal energy and space. Such a neuron will then be suitable as the basic building block for the next generation of neural networks. The research will involve the design, development and testing of the electronic neuron and subsequently a learning algorithm will be developed that can train a neural network made up of these neurons to recognise artifacts of the real world.

神经科学家现在知道，人类大脑是由数百万个小单位组成的，这些小单位被称为神经元，它们以非常复杂的方式相互连接。这些神经元使用从我们的感官（眼睛，触觉等）进入大脑的信息进行相对简单的计算，这些计算的结果作为小电信号传递给其他神经元。由于每个神经元都执行简单的计算，因此人们认为，当数百万个神经元连接在一起形成网络时，就可以实现非常复杂的计算，例如识别某人，就像人类大脑一样。工程师和科学家对大脑如何进行这些计算很感兴趣，因为大脑的计算能力远远超过任何人造机器，如台式计算机。大脑的大部分处理过程都是随着时间的推移而学习的。因此，为了了解大脑如何学习执行复杂的计算，工程师和科学家们一直在尝试建立大脑模型，称为人工神经网络。这种建模大多是使用计算机或模拟神经元行为的电子电路进行的。工程师和科学家在设计电子神经元时面临的问题是：1）设计像神经元一样工作的电路; 2）使电路足够小，以便可以将数百万个电路放置在硅芯片上; 3）这些神经元必须消耗最小的功率。由于没有可用的电子元件可以模仿神经元的元件，因此需要开发具有小物理尺寸的新电子元件，其就像真实的神经元一样操作并且消耗最小的功率。这是我们正在努力实现的目标。该项目旨在开发一种电子神经元，它具有模仿生物神经元的能力，但消耗最少的能量和空间。这样的神经元将适合作为下一代神经网络的基本构建块。该研究将涉及电子神经元的设计、开发和测试，随后将开发一种学习算法，该算法可以训练由这些神经元组成的神经网络，以识别真实的世界的伪像。