Data Lock

Inspiration

With AI changing the world on a larger scale then ever, a common concern is how we can keep data private but also train models. My solution is fl_sync which helps developers use federated learning by making communication of model parameters secure with BLE signals that are heavily encrypted, and easier by automatically converting parameters to bits to make it easy to send along with using wifi as a backup.

What it does

fl_sync helps developers use federated learning by making communication of model parameters secure with BLE signals that are heavily encrypted, and easier by automatically converting parameters to bits to make it easy to send along with using wifi as a backup.

How we built it

Used Bleak to communicate over BLE (bluetooth) and custom made encryption layers that used logarithms and signal strength to encrypt data. The primary encryption layer is AES, along with that the parameters are double ciphered. (one cipher stacked upon another).

Challenges we ran into

BLE communication, proving that Federated learning can result in better results.

Accomplishments that we're proud of

Creating a devtool you can run with any computer by pip install fl_sync.

What we learned

Using PyPI to create pip modules.

What's next for Data Lock

Better encryption and more consistent BLE connections that requires less work to setup.

Built With

• bleak
• fl-sync
• html
• pypi
• python
• pytorch
• sci-kit