Deep-Spying: Spying Using Smartwatch and Deep Learning | Hacker News Reader

Deep-Spying: Spying Using Smartwatch and Deep Learning

arxiv.org

84 pointskushti18 comments

Moneyquote of the summary: "Our results suggest that the complete technological ecosystem of a user can be compromised when a wearable wristband device is worn."

Summary:using the motion sensors in the smartwatch, they can figure out which keys you are pressing, and thus read your passwords.

I believe a recent episode of the TV series "Minority Report" used this as part of the story.

Hey guys! :)

The source code is available at: https://github.com/tonybeltramelli/Deep-Spying

And a video demo at: https://youtu.be/ZBwSfvnoq5U

A very interesting caveat from the paper:

"It is important to note that it is assumed that the victim is wearing the WAD on the wrist of the preferred hand used to interact with keyboards. In fact, in our attack scenario it would be harder, if not impossible, for the attacker to infer keystrokes if the victim is right-handed but is wearing the WAD on the left hand for example."

A WAD is any wearable arm band device.

Which is how most people.wear watches (especially expensive ones).

I don't and none of my friends seem to either.

You wear your watch on your dominant hand? That feels very unnatural to me. Just manipulating the straps alone seems difficult, so that's a barrier right there...

Not to forget, that most Watches are designed to be worn on the left side. (Position of the Buttons etc.) Smartwatches are improving here, in displaying a rotated screen, but often even these are asymmetrical designed and thus harder to use on the right arm.

> most Watches are designed to be worn on the left side.

Yes, for right-handed people. The opposite is true for left-handed people, and left-handed versions of watches.

I do as well. Even since I was a kid, wearing watches on the wrist of my non-dominant hand felt unnatural. I have no problem manipulating the strap or buttons on my watches.

That's probably a cultural thing. In my country, most people (myself included) wear watches on non-dominant hand.

I am the only right handed person that I know who wears their watch on their right hand. Most people think I am left handed because of it.

I think I started it to be different when I was a teen and the habit just stuck.

I am right-handed and wear my watch on my right wrist as well. I've never ran into any problems doing so, and manipulating buttons on different watches was never a problem either.

The reason one would want to do this is to reduce wear on the watch, since the dominant hand is going and doing things the dominant wrist is more likely to be smacked. This goes double for expensive watches.

Personally, I think it's the opposite. I wear my watch on the wrist of my dominant hand. I've never had a problem; in fact I think you are more likely to pay attention to it and not smack it into things.

Before everyone panics, there's an important note on page 70:

>The LSTM model can also successfully classify signals with an accuracy of 19% when the dataset used for training and logging are originated from two different keypads.

Still scary to think how a motivated team could extend this though.

There's an important conceptual link between training RNNs and side channel attacks.

I wrote a bit on the topic. Unfortunately, not my finest writing in terms of clarity, but I think it touches something important.

https://medium.com/@arthurb/traces-probability-and-learning-...

Reply on news.ycombinator.com