ActSemiCNNAct

This README would normally document whatever steps are necessary to get your application up and running.

Installation: Clone the code and its submodules

git clone git@github.com:HaixiaBi1982/ActSemiCNNAct.git
cd ActSemiCNNAct
git submodule update --init --recursive

This code is working with Python3.6.

python3.6 -m venv venv
source venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt

Introduction to the folders and files in this repo:

• architectures: the backbone network strcutures can be used
• data-local: the datasets which can be used for validation
• trainer: temporal ensembling based model training
• ActSemi_UCI.py: mail file for training and testing with UCIHAR dataset

Please run below lines to implement the method.

• python ActSemi_UCI.py The results will be saved to 'results' file in the root path of the repo, which will be automatically created if not existing when running the experiments.

The input files should be put at the '/data-local' directly in the format of .npy with size Ns100Nf, where Ns is the number of samples, 100 means there are 100 time points in one sample, Nf is the nubmer of features.

This method is designed to tackle the limited label issue. The main ideas of the method are as below:

• Combine active learning and semisupervised learning to select informative samples and make use of unlabelle data
• Apply a temporal ensembling-based semisupervised approach to achieve a consensus prediction using the outputs of the training networks on different epochs under network dropout regularization. (An ensemble of multiple networks generally generates more robust results than a single network)

You are greatly appreciated if you cite below paper when finding this repo useful.

@article{bi2022active, title={An active semi-supervised deep learning model for human activity recognition}, author={Bi, Haixia and Perello-Nieto, Miquel and Santos-Rodriguez, Raul and Flach, Peter and Craddock, Ian}, journal={Journal of Ambient Intelligence and Humanized Computing}, pages={1--17}, year={2022}, publisher={Springer} }