Mario Dı́az

Monica Welfert, Nathan Stromberg, Mario Diaz et al.

We propose an adversarial evaluation framework for sensitive feature inference based on minimum mean-squared error (MMSE) estimation with a finite sample size and linear predictive models. Our approach establishes theoretical lower bounds on the true MMSE of inferring sensitive features from noisy observations of other correlated features. These bounds are expressed in terms of the empirical MMSE under a restricted hypothesis class and a non-negative error term. The error term captures both the estimation error due to finite number of samples and the approximation error from using a restricted hypothesis class. For linear predictive models, we derive closed-form bounds, which are order optimal in terms of the noise variance, on the approximation error for several classes of relationships between the sensitive and non-sensitive features, including linear mappings, binary symmetric channels, and class-conditional multi-variate Gaussian distributions. We also present a new lower bound that relies on the MSE computed on a hold-out validation dataset of the MMSE estimator learned on finite-samples and a restricted hypothesis class. Through empirical evaluation, we demonstrate that our framework serves as an effective tool for MMSE-based adversarial evaluation of sensitive feature inference that balances theoretical guarantees with practical efficiency.

Shahab Asoodeh, Mario Diaz

The Noisy-SGD algorithm is widely used for privately training machine learning models. Traditional privacy analyses of this algorithm assume that the internal state is publicly revealed, resulting in privacy loss bounds that increase indefinitely with the number of iterations. However, recent findings have shown that if the internal state remains hidden, then the privacy loss might remain bounded. Nevertheless, this remarkable result heavily relies on the assumption of (strong) convexity of the loss function. It remains an important open problem to further relax this condition while proving similar convergent upper bounds on the privacy loss. In this work, we address this problem for DP-SGD, a popular variant of Noisy-SGD that incorporates gradient clipping to limit the impact of individual samples on the training process. Our findings demonstrate that the privacy loss of projected DP-SGD converges exponentially fast, without requiring convexity or smoothness assumptions on the loss function. In addition, we analyze the privacy loss of regularized (unprojected) DP-SGD. To obtain these results, we directly analyze the hockey-stick divergence between coupled stochastic processes by relying on non-linear data processing inequalities.

Shahab Asoodeh, Mario Diaz, Flavio P. Calmon

We investigate the contraction coefficients derived from strong data processing inequalities for the $E_γ$-divergence. By generalizing the celebrated Dobrushin's coefficient from total variation distance to $E_γ$-divergence, we derive a closed-form expression for the contraction of $E_γ$-divergence. This result has fundamental consequences in two privacy settings. First, it implies that local differential privacy can be equivalently expressed in terms of the contraction of $E_γ$-divergence. This equivalent formula can be used to precisely quantify the impact of local privacy in (Bayesian and minimax) estimation and hypothesis testing problems in terms of the reduction of effective sample size. Second, it leads to a new information-theoretic technique for analyzing privacy guarantees of online algorithms. In this technique, we view such algorithms as a composition of amplitude-constrained Gaussian channels and then relate their contraction coefficients under $E_γ$-divergence to the overall differential privacy guarantees. As an example, we apply our technique to derive the differential privacy parameters of gradient descent. Moreover, we also show that this framework can be tailored to batch learning algorithms that can be implemented with one pass over the training dataset.

Shahab Asoodeh, Mario Diaz, Flavio P. Calmon

We investigate the framework of privacy amplification by iteration, recently proposed by Feldman et al., from an information-theoretic lens. We demonstrate that differential privacy guarantees of iterative mappings can be determined by a direct application of contraction coefficients derived from strong data processing inequalities for $f$-divergences. In particular, by generalizing the Dobrushin's contraction coefficient for total variation distance to an $f$-divergence known as $E_γ$-divergence, we derive tighter bounds on the differential privacy parameters of the projected noisy stochastic gradient descent algorithm with hidden intermediate updates.

Tyler Sypherd, Mario Diaz, John Kevin Cava et al.

We introduce a tunable loss function called $α$-loss, parameterized by $α\in (0,\infty]$, which interpolates between the exponential loss ($α= 1/2$), the log-loss ($α= 1$), and the 0-1 loss ($α= \infty$), for the machine learning setting of classification. Theoretically, we illustrate a fundamental connection between $α$-loss and Arimoto conditional entropy, verify the classification-calibration of $α$-loss in order to demonstrate asymptotic optimality via Rademacher complexity generalization techniques, and build-upon a notion called strictly local quasi-convexity in order to quantitatively characterize the optimization landscape of $α$-loss. Practically, we perform class imbalance, robustness, and classification experiments on benchmark image datasets using convolutional-neural-networks. Our main practical conclusion is that certain tasks may benefit from tuning $α$-loss away from log-loss ($α= 1$), and to this end we provide simple heuristics for the practitioner. In particular, navigating the $α$ hyperparameter can readily provide superior model robustness to label flips ($α> 1$) and sensitivity to imbalanced classes ($α< 1$).