Differentially Private Least Squares Regression with Optimal Dimension Dependence
This work presents a sample- and time-efficient differentially private algorithm for ordinary least squares regression, with error that depends linearly on the dimension and is independent of the condition number of the design matrix.