Projects

Automatic Inference of Hyperproperties

Practical Integrity Protection for Private Computations

In this MSc thesis, we map and unify the space of existing integrity protection approaches for Fully Homomorphic Encryption to point out gaps and generic attacks. We also propose stronger and more realistic integrity notions, and explore concrete integrity primitives.

Transfer Learning in Property Inference Attacks

Predicting the privacy loss from gradient updates in collaborative learning using new affinity measures derived from transfer learning.

Veritas - Verifiable Encodings for Secure Homomorphic Analytics