Table of Contents

• About the Project
• Folder Structure
• Setup
• Findings
  • VAE: Learning Curves
  • Visualization Latent Space
  • Effect on the EEGNet
  • Effect on the EEGNet Table View

About the Project

Title: Biosignals Decoder: Improve available classifiers through data augmention using generative AI
Supervisor: Dr. Michael Knierim (IISM)
Chair: Chair of Information & Market Engineering (IISM) at Karlsruhe Institute of Technology
Date: 25.01.2024

Development of a Variational Autoencoder (VAE) to synthetically create realistic EEG data (multivariate time series). As for the data we used 10 sessions of one field study participant solving difficult math task, while 7 channels were attached to his/her brain to collect EEG data. During the field study we surveyed the mental workload that serves as our class labels (for the EEGNet). In this seminar, we have shown that by populating the training set X_train with synthetical / reconstructed data from the VAE, it stabilizes the training and decreases the validation loss and also increases the validation accuracy (see figures below).

To access the submitted seminar thesis, click here or go to the documentation dir.

Folder Structure 🗂️

📦biosignals-gen-ai
 ┣ 📂assets                    <-- Contains saved figures, ...
 ┣ 📂config                    <-- Configuration files for the pipeline
 ┣ 📂data                      <-- Provided data
 ┃ ┣ 📂raw                     <-- Contains the raw data provided by the supervisor
 ┃ ┗ 📂processed               <-- Contains the processed data
 ┣ 📂documentation             <-- Contains the PPT slides and the elaboration
 ┣ 📂models                    <-- Saved models during Development
 ┣ 📂notebooks                 <-- Jupyter Notebooks used in development
 ┃ ┣ 💻data_loader.ipynb       <-- Experimenting of data loader class
 ┃ ┣ 💻eda.ipynb               <-- Exploratory Data Analysis Notebook
 ┃ ┣ 💻eegnet.ipynb            <-- Experimenting with EEGNet and my VAE
 ┃ ┗ 💻vae.ipynb               <-- Development of my VAE (Dense)
 ┣ 📂tests                     <-- Unit tests for the source code
 ┣ 📂src                       <-- Source code / modules / classes
 ┃ ┣ 📜dataloading.py          <-- Class that handles the data loading
 ┃ ┣ 📜eegnet.py               <-- Contains the EEGNet Architecture by Lawhern et al. 2016
 ┃ ┣ 📜modelling.py            <-- Contains helper function to analyse the modelling, e.g. history plots
 ┃ ┣ 📜preprocessing.py        <-- Class that handles the data preprocessing
 ┃ ┣ 📜utils.py                <-- Contains utility / helper functions
 ┃ ┣ 📜vae_base.py             <-- Abstract class of VAE
 ┃ ┣ 📜vae_conv.py             <-- Implementation of base VAE using Conv layers
 ┃ ┣ 📜vae_dense.py            <-- Implementation of base VAE using Dense layers
 ┃ ┗ 📜vae_lstm.py             <-- Implementation of base VAE using LSTM layers
 ┣ 🕹️main.py                   <-- Entry point of the pipeline
 ┣ 📜README.md                 <-- The top-level README for developers using this project
 ┗ 📜requirements.txt          <-- The requirenments file for reproducing the environment

Setting up the environment and run the code

Note, due to time reasons, not everything is fully implemented yet. For the VAE / EEGNet, please refer to the corresponding notebooks found in .notebooks/ dir.

1. Clone the repository by running the following command in your terminal:

   git clone https://github.com/negralessio/biosignals-gen-ai
   

2. Navigate to the project root directory by running the following command in your terminal:

   cd biosignals-gen-ai
   

3. [Optional] Create a virtual environment and activate it. For example, using the built-in venv module in Python:

   python3 -m venv venv
   source venv/bin/activate
   

4. Install the required packages by running the following command in your terminal:

   pip install -r requirements.txt
   

5. Place the data in the data/raw folder.

6. Run the pipeline with the following command:

   python3 main.py --config "configs/config.yaml"
   
   

Findings

Below you can find the results of our VAE on the EEGNet.

• Populating the train set with reconstructed / synthetic data decreases the validation loss and
• also stabilizes the training, especially in the context of low samples as we have here

Learning Curves of the VAE

Visualization of the 8D latent space using PCA

Effect on the EEGNet

Effect on the EEGNet: Table View

% Synth. Added	Mean Loss	SD Loss	Mean Val Loss	SD Val Loss	Mean ACC	SD ACC	Mean Val ACC	SD Val ACC
0%	0.2318	0.0129	0.1926	0.0777	0.9273	0.0113	0.9784	0.0625
25%	0.2934	0.0094	0.1789	0.0479	0.9130	0.0087	0.9942	0.0162
50%	0.2512	0.0158	0.1518	0.0461	0.9173	0.0111	0.9986	0.0039
100%	0.2101	0.0151	0.1420	0.0475	0.9524	0.0150	0.9956	0.0164

• Above: Data Aggregation of the last epoch (#32) over all 25 runs
• Sweet Spot is 50% added reconstructed / synthetic data, as the mean validation ACC is the highest and standard deviation the lowest
• Generalization ability of EEGNet is improved
  • Significantly lower standard deviation (SD) in the validation accuracy and loss