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

Brain inspired machine learning methods for analysis of neural data

用于分析神经数据的受大脑启发的机器学习方法

基本信息

批准号：
1895488
负责人：
金额：
--
依托单位：
Imperial College London
依托单位国家：
英国
项目类别：
Studentship
财政年份：
2016
资助国家：
英国
起止时间：
2016 至无数据
项目状态：
已结题

来源：
https://gtr.ukri.org/projects?ref=studentship-1895488
关键词：
Brain inspired machine learning methods

项目摘要

Recent developments in neuroscience have produced electrical probes capable of recording the activity of hundreds or even thousands of neurons (brain cells) simultaneously. This opens the possibility for neuroscientists to discover new principles of brain function that would not have been possible with previous technology that could only record from a single neuron or a handful of neurons simultaneously. Unfortunately, despite the technology to record this data being very advanced, our ability to analyse the data has not kept pace. A number of heavily mathematical methods have been proposed. These have some promise, but share a common problem: they discover patterns in the data but they do not tell you if or how the brain itself might make use of those patterns.The aim of this research is to use brain-inspired methods to analyse the data, so that patterns that are discovered in the data would also be discoverable by the brain itself. This will lead to hypotheses that are much more closely linked to how the brain functions than existing methods.This research proposes hybrid methods that combine neural modelling with machine learning (a collection of techniques that has recently produced incredible results in solving tasks that were previously thought to require human intelligence). By incorporating neural models into the analysis, we ensure that the results are themselves something that the brain could potentially discover itself. By incorporating machine learning, we use the best currently known, state of the art methods for detecting patterns.This research has the potential to enable significant future discoveries about how the brain works and how it outperforms computers at many important tasks. It falls under the remit of the following EPSRC research areas: biological informatics; artificial intelligence technologies.

神经科学的最新发展已经产生了能够同时记录数百甚至数千个神经元（脑细胞）活动的电探针。这为神经科学家发现大脑功能的新原理提供了可能，而以前的技术只能同时记录单个或少数神经元的信息，这是不可能的。不幸的是，尽管记录这些数据的技术非常先进，但我们分析数据的能力却没有跟上。已经提出了许多高度数学化的方法。这些方法有一些希望，但有一个共同的问题：它们发现了数据中的模式，但它们并没有告诉你大脑本身是否或如何利用这些模式。这项研究的目的是使用大脑启发的方法来分析数据，这样在数据中发现的模式也会被大脑本身发现。这将导致比现有方法与大脑功能联系更紧密的假设。这项研究提出了将神经建模与机器学习（最近在解决以前被认为需要人类智能的任务方面产生了令人难以置信的结果的一系列技术）相结合的混合方法。通过将神经模型纳入分析，我们确保结果本身就是大脑可能发现的东西。通过结合机器学习，我们使用目前已知的最先进的方法来检测模式。这项研究有可能在未来发现大脑是如何工作的，以及它是如何在许多重要任务上胜过计算机的。它属于以下EPSRC研究领域的职权范围：生物信息学；人工智能技术。