mulEEG: A Multi-View Representation Learning on EEG Signals

This repostiory contains code, results and dataset links for our MICCAI-2022 paper titled mulEEG: A Multi-View Representation Learning on EEG Signals. 📝

Authors: ¹Likith Reddy, ¹Vamsi Kumar, Shivam Kumar Sharma, Kamalakar Dadi, Chiranjeevi Yarra, Bapi Raju and Srijithesh Rajendran

¹Equal contribution

More details on the paper can be found here.

Table of contents

• Introduction
• Highlights
• Results
• Dataset
• Getting started
• Getting the weights
• License and Citation

Introduction 🔥

Modeling effective representations using multiple views that positively influence each other is challenging, and the existing methods perform poorly on Electroencephalogram (EEG) signals for sleep-staging tasks. In this paper, we propose a novel multi-view self-supervised method (mulEEG) for unsupervised EEG representation learning. Our method attempts to effectively utilize the complementary information available in multiple views to learn better representations. We introduce diverse loss that further encourages complementary information across multiple views. Our method with no access to labels, beats the supervised training while outperforming multi-view baseline methods on transfer learning experiments carried out on sleep-staging tasks. We posit that our method was able to learn better representations by using complementary multi-views.

Highlights ✨

• A self-supervised model pre-trained on unlabelled Electroencephalography (EEG) data beating the supervised counterpart 💥.
• Complete pre-processing pipeline, augmentation and training scripts are available for experimentation.
• Pre-trained model weights are provided for reproducability.

Results 🕺

Linear evaluation results on Sleep-EDF dataset pre-trained on large SHHS dataset.

	Accuracy	κ	Macro F1-score
Randomly Initialized	38.68	0.1032	16.54
Single-View	76.73	0.6669	66.42
Simple Fusion	76.75	0.6658	65.78
CMC	75.84	0.6520	64.40
Supervised	77.88	0.6838	67.84
Ours	78.18	0.6869	67.88
Ours + diverse loss	78.54	0.6914	68.10

Our method performs better substantially in a low labelled data regime.

t-SNE visualization using our method (no labels used) shows clear clusters and captures the sleep-staging progression observed clinically.

Getting started 🥷

Setting up the environment

• All the development work is done using Python 3.7
• Install all the necessary dependencies using requirements.txt file. Run pip install -r requirements.txt in terminal
• Alternatively, set up the environment and train the model using the Dockerfile. Run docker build -f Dockerfile -t <image_name> .

What each file does

TODO

Training the model

TODO

Testing the model

TODO

Logs and checkpoints

• The logs are saved in logs/ directory.
• The model checkpoints are saved in checkpoints/ directory.

Getting the weights 🏋️

Download the model weights for all the baseline methods and ours.

Name	Sleep-EDF	SHHS
Single-View	link	link
Simple Fusion	link	link
CMC	link	link
Supervised	link	link
Ours + diverse loss	link	link

License and Citation 📰

The software is licensed under the Apache License 2.0. Please cite the following paper if you have used this code:

@misc{kumar2022muleeg,
    title={mulEEG: A Multi-View Representation Learning on EEG Signals},
    author={Vamsi Kumar and Likith Reddy and Shivam Kumar Sharma and Kamalakar Dadi and Chiranjeevi Yarra and Bapi S. Raju and Srijithesh Rajendran},
    year={2022},
    eprint={2204.03272},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}