This repository contains the implementation of ObscureNet and the baselines proposed in our IoTDI'21 paper entitled "ObscureNet: Learning Attribute-invariant Latent Representationfor Anonymizing Sensor Data".
Each directory is named after a privacy-preserving method described in the paper.
The 2 Human Activity Recognition (HAR) datasets used to evaluate different methods are MotionSense and MobiAct. You can download them from the following websites and use the provided converter (dataset_builder.py) to preprocess the data and turn it into the format that our code expects:
To reproduce the results of our paper, use the CSV file dataset_subjects, which is provided in this repo, instead of the original one that comes with the MobiAct dataset.
Package | Version |
---|---|
Python3 | 3.6.9 |
Tensorflow | 1.14.0 |
PyTorch | 1.4.0 |
Keras | 2.3.1 |
Omid Hajihassani, Omid Ardakanian and Hamzeh Khazaei. 2021. ObscureNet: Learning Attribute-invariant Latent Representation for Anonymizing Sensor Data, In Proceedings of the 6th ACM/IEEE Conference on Internet of Things Design and Implementation (IoTDI).