EEG to MNIST Prediction using Transformer Architecture

Project Overview

This project aims to use electroencephalogram (EEG) data to predict MNIST numbers. Identifying the power of transformer-based architectures, this project will strive to adapt it for use with EEG data. The goal is to create a robust model that can efficiently decode EEG patterns to corresponding MNIST digits.

Dataset

EEG to MNIST Dataset

For our project we will be using MindBigData's 2023 MNIST-8B dataset that contains 140,000 two second labeled data points. Each data point corresponds to a particular MNIST digit. Dataset.

• EEG data: Contains raw EEG data captured from 128 channels sampled at 250hz
• Labels: MNIST numbers 0-9 that the subject was both visualizing and listening to

Model Architecture

Transformer Based Model

Our model is based on the transformer architecture, which has shown significant success in various machine learning tasks. We adapt the transformer to handle time-series EEG data, extracting patterns and relationships that can predict the MNIST numbers.

Dependencies

• Python 3.11.x
• Pytorch
• NumPy
• MNE