User Authentication through Typing Patterns

As the digital world grows, security becomes more and more important. One specific area of interest is user authentication, which needs to be protected against impersonation and fraud. As the conventional methods such as password login begin to be vulnerable to attacks, we were curious to find other ways to verify users. Inspired by our interest in behavioral sciences, we explored ways to uniquely identify a person from their behavioral patterns, specifically their typing patterns. The goal of this project is to empower typing-based biometrics authentication, where a user may be verified based on how they type a phrase. Working towards this goal, we will implement a 1D convolutional neural network which will learn the features of the user’s typing patterns. To achieve this, the model will incorporate a meta-learning algorithm so that it is able to adapt to support new users who newly register, in effect becoming a few-shot classifier. We will test the classifier with a GAN, to generate intention hard negatives which forces the embedding and classifier to learn tight separation.

Built With

• pytorch