Machine Learning for Social EEG - A Bayesian approach to correlated component analysis and recording simultaneous multiple subject EEG
|Simon Kamronn, Andreas Trier Poulsen|
|Abstract||This thesis describes the derivation and implementation of BCoCA, a Bayesian approach to Correlated component analysis, which was introduced in Dmochowski et al. . BCoCA enables the comparisons between more than two subjects at the same time, and relaxes the constraint of equal weights with an adaptable parameter controlling the similarity between the weights for each dataset, with the purpose of locating neural activations that are synchronised within and between brains.|
The thesis outlines the principles of variational inference, the method of approximation used to derive the updates for BCoCA as well as a cost effective way to calculate its corresponding lower bound, which can be used as a measure of performance and to estimate the time of convergence. To show its capabilities BCoCA will be tested on simulated data under varying conditions, on real EEG datasets from two other experiments and will finally be used to analyse the results of an experiment conducted for this thesis.
The presented study will investigate whether neural correlations are detectable using consumer-grade hardware, with the specific goal to examine the difference between neural correlation originating from emotionally arousing and neutral films as done in Dmochowski et al. . To expand on their experimental setup and investigate the effect of experiencing an emotionally laden stimulus in a group as compared to experiencing it alone, simultaneous EEG of nine subjects were recorded. In total were 42 subjects used for the experiments.
It was shown that neural correlation is detectable using consumer-grade hardware and that it was possible to reproduce some of the results in Dmochowski et al. , showing that there is a significant difference between neural correlation originating from emotionally arousing and neutral films, respectively. The results were further established by comparing scenes with periods of significant correlation and scalp projections of the neural activity. The latter showed higher activation in areas related to emotion for the emotionally intense Sophie’s Choice compared to the suspenseful but otherwise emotionally indifferent Bang! You’re Dead. It was unfortunately not possible to determine, whether the effect of experiencing an emotionally laden stimulus in a group is significantly different to experiencing it alone. We maintain the belief that there is a difference, but further processing is needed to reveal it.
|Type||Master's thesis [Academic thesis]|
|Publisher||Technical University of Denmark, Department of Applied Mathematics and Computer Science|
|Address||Matematiktorvet, Building 303B, DK-2800 Kgs. Lyngby, Denmark, email@example.com|
|Series||DTU Compute M.Sc.-2013|
|Note||DTU Supervisor: Lars Kai Hansen, firstname.lastname@example.org, DTU Compute|
|BibTeX data|| [bibtex]|
|IMM Group(s)||Intelligent Signal Processing|
Back :: IMM Publications