… of Empirical Risk Minimization (ERM) with (smooth) non-convex loss functions under the
differential-privacy (… We first study the expected excess empirical (or population) risk, which was …

… property of the private estimator. By using l2 norm of the gradient in the empirical risk, we
show an upper bound of the population risk with non-convex loss functions at the point θpriv …

… differentially private Empirical Risk Minimization (ERM) problem in different settings. For
smooth (strongly) convex loss function … empirical risk from convex loss functions to non-convex …

… with DP-ERM and smooth convex loss functions in high … non-convex loss functions, we
explore both low and high dimensional spaces. In the low dimensional case with a non-smooth …

… Differentially private empirical risk minimization with smooth non-convex loss functions: A
non-stationary view. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33…

… In this paper, we study differentially private Empirical Risk Minimization in the noninteractive …
holds for non-convex loss functions and constrained domain set, as long as they are smooth …

… nonnegative loss function ℓ … losses to be smooth and convex. While these are common
assumptions adopted in the analysis of private ERM [29, 10], the generalization to the non-convex …

… For self-completeness, we will rephrase them so that they fit our differentially private context.
… Before doing that, we show that the empirical risk is Lipschitz and smooth, which satisfies …

… The empirical risk of any θ ∈ Rd wrt loss ℓ and dataset D := {… Private empirical risk minimization:
Efficient algorithms and … New results in convex and non-convex settings. Advances in …

… loss functions with convex constraints arise both directly (eg, because of a convex log-likelihood
function) or as proxies for more complicated (non-convex… sequence of soft constraints by …