DP-GD achieves minimax optimal non-asymptotic risk O(γ + γ²/ρ²) for well-conditioned high-dimensional data and power-law scaling for ill-conditioned power-law spectra, with the exponent depending on the privacy parameter ρ.
Privacy for Free in the Overparameterized Regime,
2 Pith papers cite this work, alongside 1 external citations. Polarity classification is still indexing.
2
Pith papers citing it
1
external citations · OpenAlex
representative citing papers
Optimal codeword arrangement in Hamming codes for BSC transmission of counting queries strictly improves differential privacy without added obfuscation or utility loss.
citing papers explorer
-
High-Dimensional Private Linear Regression with Optimal Rates
DP-GD achieves minimax optimal non-asymptotic risk O(γ + γ²/ρ²) for well-conditioned high-dimensional data and power-law scaling for ill-conditioned power-law spectra, with the exponent depending on the privacy parameter ρ.
-
Differential Privacy over Hamming Codes
Optimal codeword arrangement in Hamming codes for BSC transmission of counting queries strictly improves differential privacy without added obfuscation or utility loss.