Publikace na Matematicko-fyzikální fakulta |
2022
Abstrakt
In this paper, we focus on the problem of user's authentication through typing dynamics patterns. We specifically focus on small-sized problems, where it is difficult to fully train corresponding machine (deep) learning algorithms from scratch.
Instead, we propose a different approach based on the visualization of the typing patterns and subsequent usage of pre-trained feature extractors from the computer vision domain. We evaluated the approach on a publicly-available dataset and results indicate that this is a viable solution capable to improve over several baselines.
Moreover, the proposed visual representation of the data contributes to the explainability of AI.