Data Reduction Techniques for Systematic Information Quantification in Large Scale, Multiple Spike Trains

人们普遍认为，大脑的功能在很大程度上取决于位于不同大脑区域的大量不同神经元群之间的相互作用。为了从经验上测试这些神经元群是如何协同工作以产生感觉和感知等功能的，神经科学家同时记录了大脑中不同神经元群的活动。该活动的记录可以通过许多不同的方法来实现。多电极已成为研究特定脑区或不同脑区的多个神经元同时活动的标准工具。编码在脉冲序列中的刺激信息是神经科学研究的主要焦点，并且经常根据各种反应或特征进行检查，例如脉冲计数，平均反应时间，第一脉冲潜伏期，以及脉冲间隔和放电率。在越来越多的实验中记录的大量响应数据阵列对数据分析和神经元功能的解释和建模提出了许多挑战。在这项可行性研究中，我们建议将系统工程、数据约简技术和多元统计中的信息理论专业知识结合起来，以提供一种最有效地分析大型神经群体信息内容的方法。特别地，我们将尖峰序列视为一个信息编码和解码过程，并提出在信息理论框架下使用小波、ICA和地形映射来提取、分类、量化和组织尖峰序列中包含的信息特征；我们建议使用自组织数据约简技术来改进互信息的估计，并探索许多不同的方法来提供适合于不同性质的大脑信号组合分析的信息理论方法。

It is widely believed that the function of the brain crucially depends on the interaction between large numbers of different neuronal populations located in different brain areas. To test empirically how these neuronal populations work together to generate functions such as sensation and perception, neuroscientists record simultaneously the activity of different neuronal populations in the brain. Recording of this activity is achieved by a number of different methods. Multi-electrodes have become a standard tool for studying the simultaneous activity of multiple neurons in a specific brain region or across different regions. The stimulus information encoded in the spike trains is a primary focus of research in neuroscience and is often examined in terms of various responses or features such as spike counts, mean response time, first spike latency, as well as interspike intervals and firing rate. The massive response data arrays recorded in growing number of experiments pose many challenges for data analysis and for interpreting and modelling of neuronal functions. In this feasibility study we propose to bring together the information theoretic expertises in the systems engineering, data-reduction techniques and the multivariate statistics in order to provide a method to analyse most effectively the information content of large neural populations. In particular, we consider spike trains as an information encoding and decoding process and propose to use wavelets, ICA and topographical mappings to extract, classify, quantify and organise information features contained in spike trains under the information theoretic framework; and we propose to use self-organised data reduction techniques for improved estimation of mutual information, and also to explore a number of different approaches to provide information-theoretic methods suited to the combined analysis of brain signals of different nature.