1 of 10

Real-time machine learning-based user authentication via daily activities using wireless signals

​

​

Advisor: Prof. Yingying Chen

2 of 10

Hello!

Name: Bhargav Singaraju

Major: Electrical Engineering

Fun Fact: My favorite candy is Kit Kat.

​

​

2

3 of 10

Hello!

Name: Rishika Sakhuja

Major: Computer Engineering

Fun Fact: I have collected over 100 keychains from all of the places I have travelled to.

3

4 of 10

Hello!

Name: Sachin Mathew

Major: Computer Engineering and Computer Science

Fun Fact: In high school I ate gallon of sorbet in under an hour and I my body will never be the same.

4

5 of 10

CSI Preprocessing

  • Processed a DAT file (walking) with CSI
  • Visualized output from 30 different subcarriers
  • Output will be used to identify the user’s activity
  • Simple Sequence - time sequence
  • Amplitude - Useful for activity recognition and identifying users based on large scale movements

5

6 of 10

6

7 of 10

Real Time Segmentation Prereq

  • CSI writes 1000 packets of data per second
  • Collect last second of data and pass it into DNN
  • Even if the desired action is longer than a second long, DNN classifies data based on current datapoint as well as the last datapoint passed into the DNN

​

​

​

7

8 of 10

Visualization Tool

  • Bokeh- interactive data visualization tool to display real time CSI data
    • Display subcarriers
  • Use python to connect to Bokeh and graph in real time
  • Line plots to show spikes when the user is performing an activity

​

​

​

​

8

9 of 10

Next Steps

​

  • Load real-time CSI to TensorFlow models
  • Use TensorFlow to implement real-time data segmentation mechanism
  • Continue developing visualization tool
  • Collect CSI data for various activities

​

​

9

10 of 10

Questions?

10