We study the limits and capability of public-data assisted differentially private (P A-DP) algorithms. Specifically, we focus on the problem of stochastic convex optimization (SCO) with either labeled or unlabeled public data.

Submodular maximization has become established as the method of choice for the task of selecting representative anddiversesummaries of data.

This models applications where learners crowdsource information from non-expert human workers or conduct noisy experiments to determine group structure. The learner aims to exactly recover a partition by submitting queries of the form "are

Notably,we find that neuronal stochasticity plays a large role in the white box adversarial robustness, but that stochasticity alone is insufficient to explain ourresults--neuronal stochasticity interactssupralinearly with theVOneBlock features to driveadversarialrobustness.

Inmanycases,thenumber ofdimensions in the data is much larger than the number of actual data samples.