Generative modeling applied to EEG for affect detection using commercially available hardware.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data
eeg-ml
images
README.md

README.md

Machine Learning for EEG (Electroencephalography) Affect Identification

Background

This is the central repository for the affect identification project of BrainLab @ the Georgia Tech School of Interactive Computing. The goal of this project was to utilize various available machine learning techniques to tackle the problem of affect identification in humans from EEG using commercially available hardware.

As a Classification Problem

Statistical Classification attempts to address the problem of identifying, given a sample from an unknown distribution X, the most likely category C it corresponds to. In our case, this means attempting to determine the most probable state of an individual given a recent window of their EEG.

As a Generative Modeling Problem

Generative Modeling in contrast actually attempts to learn the entire underlying distribution of a set of samples itself. With respect to our goals, this means attempting to generate a realistic looking sample of a person's EEG given their current state.

This can actually be used, through sampling, to address the classification problem of determining the conditional probability of a given state using EEG. Although, you also get the benefit of being able to draw an arbitrarily high number of samples from a minniature version of the true underlying distribution, assuming your generative model has converged correctly. (Tough assumption.)