More to Less (M2L): Enhanced Health Recognition in the Wild with Reduced Modality of Wearable Sensors

This is the Pytorch implementation of the paper More to Less (M2L): Enhanced Health Recognition in the Wild with Reduced Modality of Wearable Sensors

Motivation.

• Wearable sensors are well suitable for continuous and ubiquitous monitoring of individual’s physiological and behavioral profiles in the free-living context (unobtrusive, affordable and user-friendly).
• Multiple sensors can provide different measurements of the subjects (i.e., chest pad, wrist wearable sensor, EEG sensors), and the different measurements usually contain complementary information, which is beneficial for healthcare applications.
• Therefore, fusing information from multiple sensors is a common choice in many applications

However, most works is built on the assumption: have access to an equal number of sensors in both training and testing

Figure 1. Motivation.

• The assumption does not hold true in many real-world scenarios due to various factors, i.e., Device size/cost, Energy consumption constraint, Signal lost, User’s preference, etc.
• Therefore, it is critical to bridge the gap between the models developed using multiple sensors during development and the models using less number of sensors during deployment in the wild.

Requirements

• Python 3.7
• numpy
• Pytorch == 1.4.0 and later
• scikit_learn

Folders

The code can be applied to other dataset as well, only if the data is organized by the required format.

• ./data: The required data files should be put into
• ./configs: place to modify the hyper-parameters used in the project
• ./experiments: save the model's parameters and hyper-parameters into this folder.
• ./models: model's definition
• ./utils: utility functions

Data

Right now the model is tested on both SMILE and TILES dataset.

Usage

Commands for training and testing the model.

# provied the model name you'd like to use in the opts.py file.
python ./train.py

The detailed descriptions about the arguments are as following:

Parameter name	Description of parameter
DATA_PATH	folder of the data file.
MODEL_PATH	folder to save model and other results.
CUDA_DEVICE	the GPU device you want to use, otherwise, the model will be run on CPU
NUM_CLASS	number of classes, for example 2 for binary classification (one-hot coding used)
BATCH_SIZE	batch size
TRAINING_EPOCHES	how many epoches you'd like to use for the training
LEARNING_RATE	learning rate, i.e., 0.001
DECAY_RATE	adjust learning rate with training epoch, which is used with lr_scheduler.
STEP_SIZE	adjust learning rate every STEP_SIZE by DECAY_RATE.
INPUT_LEN	Input sequence length of (for example, 7 days)
INPUT_DIM_M1	dimension of modality ONE
INPUT_DIM_M2	dimension of modality TWO
EMBED_DIM	dimension of embeded size.

Results

Citation

If you use the code, please cite our work:

@inproceedings{yang2022more,
  title={More to Less (M2L): Enhanced Health Recognition in the Wild with Reduced Modality of Wearable Sensors},
  author={Yang, Huiyuan and Yu, Han and Sridhar, Kusha, etc },
  booktitle={2022 44st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)},
  year={2022},
  organization={IEEE}
}