Skip to content
 
 

Repository files navigation

AIDL - EEG Based Emotion Recognition

Summary

This project seeks to apply deep learning techniques to electroencephalography (EEG) data collected in the context of subject emotion recognition.

On the SEED-IV database, we utilized a variety of modern deep learning approaches including high-dimensional convolutional networks as well as hybrid convolutional and recurrent models. Utilizing this convolutional approach on preprocessed data achieved state of the art results at 73% accuracy utilizing the LOSO (leave one subject out) experimental format.

We conclude that models which adequately encompass the concept of locality across multiple spatial, temporal, and frequency domains have the best performance characteristics.

Instructions

Requirements

Docker 🐳 19.03+ is required. A container is used to manage other requirements.

Data

Datasets currently used in this project are SEED-IV and OpenNEURO. In order to replicate the results of the code in this repository, acquiring permission and downloading the datasets are linking them to the respective ./data repository of each dataset is necessary.

Running Code

Models are implemented using tensorflow. Replication of results can be acheived by running the following docker script. This script will build a docker container containing project dependencies if not already built locally. The docker container follows the nightly release of tensorflow with GPU and jupyter notebook support. If running on a machine with correct NVIDIA drivers, models will be trained with GPU acceleration. The script will launch a Jupyter Notebook in the container which contains replicable project code and results for each dataset in respective .ipynb notebook files.

Current Status

The SEED-IV dataset is the dataset being primarily studied. Previous, but currently incomplete work on the ds003004 dataset is also contained. Further datasets may be added.

Credit

Authors

This project is the work of Atneya Nair and Akum Kang. Unless otherwise noted, contributions to this repository represent their equal work.

Institution

This repository is a subproject of the AI-based Discovery and Innovation vertically integrated project at Georgia Tech, led by Prof. Ali Abidi.

About

This project seeks to apply deep learning techniques to electroencephalography (EEG) data collected in the context of subject emotion recognition.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages