The paper notes that these types of attacks can be carried out with off-the-shelf equipment and algorithms.
This has basically zero practical application.
The training data knew which keys were being pressed repeatedly to allow identification.
So you need physical access to the keyboard to train the system. You cannot just listen in to a person typing on an unknown keyboard and know the keystrokes.
If you have physical access…a keylogger will do the job with much less effort.
That’s a relief in a sense, I guess.
Thanks. Damn that’s such a stupid headline
True. But who would read a story with the headline:
“Research team create learning model with zero real world applications.”
The paper notes that while completely useless, it can be done with off the shelf equipment…but it is still cheaper and actually effective to use a $10 keylogger.
I’d read that. Because the honest title, and I’d be curious as to why they’re doing it, and if there might be a niche way to actually make it worth it one day.
So you need physical access to the keyboard to train the system.
You need physical access to a keyboard to train the system. Knowing what keyboard your target uses seems easier than gaining physical access to it, assuming you want to stay undetected.
I can imagine one way to do this, but it’s a bit of an edge case; if you can get into a video chat with the person and then get them to DM you stuff while on the line, you’d be able to get a recording of their keyboard sounds along with the text it produced.
There’s probably easier and more reliable ways to compromise someone, though.
