Learning while Respecting Privacy and Robustness to Distributional Uncertainties and Adversarial Data

Abstract

Data used to train machine learning models can be adversarial--maliciously constructed by adversaries to fool the model. Challenge also arises by privacy, confidentiality, or due to legal constraints when data are geographically gathered and stored across multiple learners, some of which may hold even an "anonymized" or unreliable dataset. In this context, the distributionally robust optimization framework is considered for training a parametric model, both in centralized and federated learning settings. The objective is to endow the trained model with robustness against adversarially manipulated input data, or, distributional uncertainties, such as mismatches between training and testing data distributions, or among datasets stored at different workers. To this aim, the data distribution is assumed unknown, and lies within a Wasserstein ball centered around the empirical data distribution. This robust learning task entails an infinite-dimensional optimization problem, which is challenging. Leveraging a strong duality result, a surrogate is obtained, for which three stochastic primal-dual algorithms are developed: i) stochastic proximal gradient descent with an $ε$-accurate oracle, which invokes an oracle to solve the convex sub-problems; ii) stochastic proximal gradient descent-ascent, which approximates the solution of the convex sub-problems via a single gradient ascent step; and, iii) a distributionally robust federated learning algorithm, which solves the sub-problems locally at different workers where data are stored. Compared to the empirical risk minimization and federated learning methods, the proposed algorithms offer robustness with little computation overhead. Numerical tests using image datasets showcase the merits of the proposed algorithms under several existing adversarial attacks and distributional uncertainties.