[Mobile User Authentication Using Deep Learning]
Achieve user authentication by applying deep learning techniques to the wireless signals generated by mobile devices
Project mission
Use of WiFi signals to capture inherited behavioral characteristics to facilitate identification/authentication.
Environment Dependent Solution (Robust to Placement).
To examine CSI of WiFi to study behavioral characteristics.
Develop a CNN model resilient to spoofing attacks.
The Team
Aditi Satish
Daniel Liu
Sharad Prasad
Emily Gao
David Man
WEEKLY PROGRESS
Week 1
- Read the MASS paper from Winlab that uses deep learning techniques to perform user authentication
- Weekly Presentation Link
Week 2
- Read the MobiCom paper from our lab that uses wireless signals to perform activity recognition
- Learnt programming tools for deep learning. (Python and Tensorflow).
- Weekly Presentation Link
Week 3
- Overview of Tensorflow tools- followed the tutorial.
- Tutorial used the Fashion MNIST dataset, which has 70,000 grayscale images in 10 categories. Each item was shown in a low resolution image of 28x28 Pixels.
- Reviewed MobiCom Paper "E-eyes: Device-free Location-oriented Activity Identification Using Fine-grained WiFi Signatures".
- Weekly Presentation Link
Week 4
- Continued to familiarize ourselves with the Python packages (Tensorflow, Scipy).
- Obtained CSI data and began preprocessing.
Week 5
- Study the CSI data.
- Preprocess Wifi CSI data (denoising and segmentation) with SciPy.
- Study ways to segment CSI data using moving variance.
Week 6
- Continued processing the CSI data with SciPy and Pandas library.
- Begin programming the CNN and deep learning model with Tensorflow
Week 7
- Finish segmenting activities from CSI data.
- Record CSI data of different activities/behaviors (squatting, sitting).
- Convert time series data to frequency domain utilizing Fourier transform.
- Continue tuning feature extraction model.
Week 8
- Collect more data to be used as inputs for models (different activities, multiple users and antennas).
- Implement user and activity recognizer models.
- Continue developing & hyperparameter tuning the deep learning to boost validation accuracy
Week 9
- Finish developing & hyperparameter tuning the deep learning to boost validation accuracy
- Finish developing project page
- Complete Open House Presentation
Week 10
Code
Data Segmentation: https://colab.research.google.com/drive/16djK6L4wWOSM1DxzcM62u_KcBZ35M0Vn#scrollTo=rAQAaoh7pX1Y
Deep Learning Model: https://colab.research.google.com/drive/1RpMG24kZFY3AqP6hpyGlnXQthnBcdiTI?usp=sharing
Spectrogram: https://colab.research.google.com/drive/11ea81W1FDMeg7oiNGX3EH71Mk8wZZbr2?usp=sharing