REST : Robust and Efficient Sleep Monitoring in the Wild

Code accompanying our WWW 2020 paper.

Contents

1. Motivation
2. Setup
3. Experiments

Motivation

In the paper, we identify the following key challenges for deploying deep neural network for sleep staging in the wild. The DNN needs to be:-

C1. Robust to noise in input data.

C2. Energy Efficient. 

To tackle these challenges, we propose a novel framework - REST that builds noise robustness through adversarial training and spectral regularization and energy efficiency through sparsity regularization. The method is summarized below

(from left) When a noisy EEG signal belonging to the REM (rapid eye movement) sleep stage enters a traditional neural network which is vulnerable to noise, it gets wrongly classified as a Wake sleep stage. On the other hand, the same signal is correctly classified as the REM sleep stage by the REST model which is both robust and sparse. (From right) REST is a three step process involving (1) training the model with adversarial training, spectral regularization and sparsity regularization (2) pruning the model and (3) re-training the compact model.

Requirements

This project is written in python 3.6.6

We suggest recreating the experimental environment using Anaconda through the following steps.

1. Clone the project

   git clone https://github.com/sleepstagingrest/REST.git

2. Install the appropriate version for Anaconda from here - https://www.anaconda.com/distribution/

3. Create a new conda environment named "rest" and install requirements.

   • conda create -n rest python=3.6.6
   • conda activate rest
   • pip install --user --requirement requirements.txt

4. Install advertorch

   pip install -e advertorch

Experiments

For demo, we use the Sleep-EDF dataset

1. Prepare dataset

   • Follow instructions within data/physionet to prepare Sleep-EDF dataset

2. To evaluate Sors Model on Physionet, execute the following steps within experiments.py

   • Ensure correct config file is specified : args=args_sors_physionet
   • Specify a logging comment : args['logging_comment'] = Sors_Physionet
   • Specify the correct experiment : ex.bt()

3. To evaluate REST(A+S) model on Physionet, execute the following steps within experiments.py

   • Ensure correct config file is specified : args=args_sors_physionet
   • Specify a logging comment : args['logging_comment'] = Rest_Physionet
   • Specify the correct experiment : ex.atc_ortho()

4. Each experiment in steps 2,3 produces a log file which can be viewed to get correct result.

   • The path of the file is logs1/[dataset]_[model]/['logging_comment'] where
     • dataset is physionet
     • Model is Sors
     • logging_comment is Sors_Physionet or Rest_Physionet depending on experiment

License

This code is released under the MIT License (refer to the LICENSE file for details).