This title appears in the Scientific Report :
2015
Please use the identifier:
http://hdl.handle.net/2128/9646 in citations.
Comparative correlation analyses of high-dimensional point processes: applications to neuroscience
Comparative correlation analyses of high-dimensional point processes: applications to neuroscience
Modern electrophysiological techniques allows to record the electrical extra-cellular potential from the cerebral cortex simultaneously from 100 or more electrodes. This makes possible to reconstruct the electrical impulses (“spike trains”) of tens or hundreds of neurons in parallel. Such massivel...
Saved in:
Personal Name(s): | Quaglio, Pietro (Corresponding Author) |
---|---|
Contributing Institute: |
Computational and Systems Neuroscience; INM-6 Computational and Systems Neuroscience; IAS-6 |
Published in: | 2015 |
Imprint: |
2015
|
Physical Description: |
93 |
Dissertation Note: |
Universita di Torino, Masterarbeit, 2015 |
Document Type: |
Master Thesis |
Research Program: |
Connectivity and Activity |
Subject (ZB): | |
Link: |
OpenAccess OpenAccess |
Publikationsportal JuSER |
Modern electrophysiological techniques allows to record the electrical extra-cellular potential from the cerebral cortex simultaneously from 100 or more electrodes. This makes possible to reconstruct the electrical impulses (“spike trains”) of tens or hundreds of neurons in parallel. Such massively parallel data led to the development of several statistical methods for detection and evaluation of higher-order correlation, that are supposed to play a fundamental role in network processing and coding in the brain.There exist a number of statistical methods that are able to detect and identify different correlation structures in such massively parallel data. The goal of this work is to compare some of these methods by applying them to the same (simulated) data of known ground truth. The results obtained by application on the same data designed with different correlation structures enlighten different aspects of correlation structures which are often difficult to relate to each other. In this thesis we consider two methods: the CuBic method (Staude et al. 2009) and SPADE (Torre et al. 2013). The first is designed to extract a lower bound of exisiting higher-order correlations, the other extracts synchrony patterns occurring significantly. We propose a general approach for testing and comparing the different analysis tools. The integration of the results from the different methods is fundamental for a better interpretation of the correlation structure underlying the analyzed data.We proceeded by presenting a framework to model and simulate generic correlated spike trains. Then we applied the two methods to such simulated data and critically interpreted the results. We showed that the integrated information from the results of the two methods allow to identify the actual correlation structure in most of the simulated datasets.We propose the use of simulated datasets as a general procedure to test and calibrate statistical analyses.Eventually we focused on parameters the generative model for which a direct comparison of the different methods' result is not possible. We analyzed in detail the performance of the methods for this particular parameter settings for large number of realizations of simulated data. |