Hi HN community,
My name is Zacharie Rodière (emphasis on the accent), a recent MS ECE graduate from Georgia Tech, originally from France. Since my graduation, I decided to work on a startup in the field of continuous authentication and behavioral AI. The name of my company is BehavLabs.
I recently released our first continuous auth demo using keystroke biometrics. The model is trained on mostly open-access data but I can't much discuss the model architecture here (if any ML engineers want to discuss it privately feel free to reach out).
The demo consists of two modes: a two-player mode where two people face off and the model tries to guess if the people are different or the same, and a 3+ player mode with 'compare' and 'detect' modalities. In compare mode you get a similarity matrix showing the model's similarity score for each user pair. In detect mode, each user types a prompted excerpt, then one of the n users types a final prompt, and the model tries to guess who typed the final prompt.
I'd say the model works pretty well even if we're not at 100% accuracy yet. I'd love y'alls feedback on it, whether it's about the ML, UI or security front.
Comments URL: [https://news.ycombinator.com/item?id=49353290](https://news.ycombinator.com/item?id=49353290)
Points: 1