Neurotech UCB

Our Projects

Deep Learning for Affective State Recognition during Gameplay on the GAMEEMO Dataset

Project code on Google Colab

Highlights of the Work:

The Database provides EEG-based emotion data with respect to computer games.
EEG data was collected with a wearable and portable device that differs from the traditional approaches.
The authors of the paper demonstrated the performance of the wearable EEG device in comparison with the traditional ones.
Showed the success of computer games in contrast to the other aural-visual stimuli methods composed of video clips and short movies.
We segmented the signal in many pieces and used Wavelets to create scalograms of the EEG.
We used a type of CNN (Convolutional Neural Network) known as Inception Network to classify emotions.

The EEG channels used. Green channels are reference electrodes.

The arousal-valence model used to measure emotion.

Scheme of a Inception Neural Network.

On the arousal-valence model, the first area includes high arousal positive valence (HAPV) emotions ranging from pleased to excited. Area 2 ranges from nervous to annoying emotions called high arousal negative valence (HANV). Area 3 consists of three negative emotions (sad, bored and sleepy) and can be represented as low arousal negative valence (LANV). The last area is called low arousal positive valence (LAPV) which includes calm to relaxed emotions.

Other Useful Resources:

Gameemo Dataset Dataset

Gameemo paper: Database for an emotion recognition system based on EEG signals and various computer games

InceptionV3 Neural network article: The Transfer/Domain/Few Shot Learning Model we used

Inception Nets Intro: Inception Nets Theory

Learn Keras with examples: Keras Example Codes

Friendly Intro to Wavelets: Wavelets and CNNs

A Background on the emotion classification problem: What is Emotion Classification?