Neural Network Typing Authentication

1
Neural Network Typing Authentication

• TJ STAR
• Luke Knepper
• lukeknep_at_stanford.edu

2
The Dilemma

• Passwords can sometimes be suboptimal
• Advanced biometrics are expensive
• Need an alternative

3
A Solution

• Authenticate people by how they type
• Typing patterns differ by person ? users
  specific typing characteristics will be used for
  authentication
• Cheap and flexible to implement

4
(No Transcript)
5
Background

• Measures users' typing patterns, compares to a
  previous standard
• Technique first used in WWII
• Works with 90 Accuracy
• Effectively implemented in a neural network
  structure

6
Background
7
Development (front-end)

• On account set-up, user types large amounts of
  text
• On subsequent log-ins, user types smaller amount
  of text
• Username and password are still needed

8
Development (back-end)

• Set-up data is used to train a neural network
• The node weights are tweaked thousands of times
  to train the neural network to output the desired
  result
• Log-in data will be fed through neural network
  result either meets threshold (admitted) or does
  not meet (rejected)

9
Experimentation

• Step 1
• Collected lots of test typing data via an
  online applet
• www.lukeknepper.com/research.html

10
Experimentation

• Step 2
• Use this data to test the accuracy of the neural
  network approach
• Train a neural network for each data file
  collected, run every other file through that
  neural network, record how many breach

11
Experimentation

• Step 3
• Tweak the neural network set-up and repeat step
  2 to find the optimal neural network

12
Results
Network Type Mean Training Time Breached (threshold of .007)
Base Network 1.1 s 90.5
Increased exposure (25 ? 50) 2.3 s 90.7
Increased Cycles (1000 ? 2000) 2.7 s 90.9
Increasing Layers (1 ? 3) 3.1 s 90.5
Increasing All 7.5 s 90.9
13
Another Advantage

• What if another person jumps on the computer
  while you are logged in?
• Can continuously monitor the user's typing
  patterns during program use
• If a change is detected, system suspects an
  intruder and locks the user out

14
Demo

• Lets have some fun.

15
(No Transcript)
16
Neural Network Typing Authentication

• TJ STAR
• Luke Knepper
• lukeknep_at_stanford.edu