nann-cheng/condeval

A framework for secure cosine-simliarity computation via the CondEval primitive

This is a fast proto-type implementation of the protocol in the paper Privacy-preserving Cosine Similarity Computation with Malicious Security Applied to Biometric Authentication.

Offline phase

We have the bank distribute secret sharing of the required correlated randomness in the offline phase, which happens before online authentication, generate correlated randomness data written to data/offline.pk0 and data/offline.pk1.

python3 dealer.py

Online phase

The protocol run in a setting with four parties.

• A client submitting secret input to the two servers.
• A bank generating offline phase correlated randomness as well as referenced template for authentication.
• Two servers responsible for executing secure computation.

To test the protocol, sequentially run

python server.py 0 python server.py 1