Differentially Private Bilevel Optimization: Efficient Algorithms with Near-Optimal Rates

Daogao Liu (University of Washington, Seattle) · Andrew Lowy (University of Wisconsin - Madison)
approximate differential privacyapproximate stationary pointsbilevel optimizationconvex objectivedifferentially private optimizationempirical riskexponential mechanismhyperparameter optimizationinexact function evaluationslog-concave samplingmeta-learningnon-convex algorithmspure differential privacyregularized exponential mechanismstochastic optimization

Bilevel optimization, in which one optimization problem is nested inside another, underlies many machine learning applications with a hierarchical structure