Secure authentication based on deep learning

The activity aims at exploiting adversarial neural networks to perform biometric user authentication. It involves a significant paradigm shift with respect to the state of the art, whereby features are not only learned from training data, but they are organized in the latent space in a way that makes the authentication process more accurate. The activity also involves obfuscating part of the information in order to prevent attacks based on eavesdropping.

ERC Sector:

  • PE7_7 Signal processing


  • Authentication
  • Artificial neural networks

Research groups