Search papers, labs, and topics across Lattice.
2
0
3
Forget choosing between differential privacy and homomorphic encryption in federated learning – this round-based interleaving strategy dynamically combines them with synthetic data for tunable privacy-quality-efficiency trade-offs.
By cleverly interleaving homomorphic encryption with synthetic data training in federated learning, this work achieves a sweet spot: better accuracy *and* lower encryption costs.
