Machine learning/artificial intelligence is an emerging field, and we wanted to apply this profound technology to an everyday use case - the smartphone. Privacy and security are especially important in the 21st century, and we wanted to apply machine learning fundamentals to learn a user's pattern unlocking tendencies and be able to detect intruders who may know the phone's unlock pattern.
What it does
The lockscreen is equipped with a neural network which can differentiate between the phone's owner and an intruder trying to unlock it. The initial set-up involves the unlock pattern being repeatedly drawn by both the owner and an outside individual so the neural network is able to detect subtle differences between users based on a variety of criteria. The AI is further trained and improved each time it is unlocked, either by the owner or a detected intruder.
How we built it
Neural network architecture was constructed, measuring various criteria and evaluating various mathematical operations based on these findings to build a profile of the phone's owner.
Challenges we ran into
Unable to get neural network's propagation algorithm to converge at a low enough error, overriding the home button on Android to get an actual lock screen.
Accomplishments that we're proud of
Building an efficient neural network that is up to 90% successful in our performed test cases.
What we learned
The intricacies of artificial intelligence; small operational changes can have large-scale impacts on the success rate.
What's next for LearningLock
Integrating additional functions to improve user experience (ex. widgets showing content relevant to the user).