Search papers, labs, and topics across Lattice.
1
0
3
Differential privacy in language tasks is surprisingly cheap: approximate DP is free, and pure DP only reduces performance by a factor of $\min\{1,\varepsilon\}$.