/ODIN

Domain Adaptation with Representation Learning and Nonlinear Relation for Time Series

Primary LanguageJupyter NotebookMIT LicenseMIT

ODIN

This repository contains the code used for the paper titled Domain Adaptation with Representation Learning and Nonlinear Relation for Time Series" by Hussein A., Hajj H.

High level approach description

Alt text

Model

Alt text

Requirements

  • python version 3.6.12
  • To create anaconda environment run conda env create -f environment.yml

Quick start

  1. Download the PAR/HARR datasets
  2. Run main.py in one of the following modes:
    • cr_user: cross user
    • cr_device: cross device
    • cr_user_device: cross user and cross device
python main.py --dataset PAR --mode cr_user --path "path/to/dataset"

Sample of reconstructed signals from test set after adaptation

  • Run inspect_AE.py to generate sample figures of advesarial examples Alt text Alt text

Domain adaptation toy example

Toy examples for the limitations of domain adaptation with hard parameter sharing and how domain adaptation with soft parameter sharing overcomes these limitations Open In Colab

Alt text

Alt text

Alt text

Contacts

Cite Paper:

@article{hussein2022domain,
  title={Domain Adaptation with Representation Learning and Nonlinear Relation for Time Series},
  author={Hussein, Amir and Hajj, Hazem},
  journal={ACM Transactions on Internet of Things},
  volume={3},
  number={2},
  pages={1--26},
  year={2022},
  publisher={ACM New York, NY}
}