Privacy Preserving Recalibration under Domain Shift

by   Rachel Luo, et al.

Classifiers deployed in high-stakes real-world applications must output calibrated confidence scores, i.e. their predicted probabilities should reflect empirical frequencies. Recalibration algorithms can greatly improve a model's probability estimates; however, existing algorithms are not applicable in real-world situations where the test data follows a different distribution from the training data, and privacy preservation is paramount (e.g. protecting patient records). We introduce a framework that abstracts out the properties of recalibration problems under differential privacy constraints. This framework allows us to adapt existing recalibration algorithms to satisfy differential privacy while remaining effective for domain-shift situations. Guided by our framework, we also design a novel recalibration algorithm, accuracy temperature scaling, that outperforms prior work on private datasets. In an extensive empirical study, we find that our algorithm improves calibration on domain-shift benchmarks under the constraints of differential privacy. On the 15 highest severity perturbations of the ImageNet-C dataset, our method achieves a median ECE of 0.029, over 2x better than the next best recalibration method and almost 5x better than without recalibration.


page 1

page 2

page 3

page 4


Correlated Data in Differential Privacy: Definition and Analysis

Differential privacy is a rigorous mathematical framework for evaluating...

Discrete Distribution Estimation with Local Differential Privacy: A Comparative Analysis

Local differential privacy is a promising privacy-preserving model for s...

Privacy-Preserving Synthetic Datasets Over Weakly Constrained Domains

Techniques to deliver privacy-preserving synthetic datasets take a sensi...

Privacy Leakage over Dependent Attributes in One-Sided Differential Privacy

Providing a provable privacy guarantees while maintaining the utility of...

Renyi Differentially Private ADMM Based L1 Regularized Classification

In this paper we present two new algorithms, to solve the L1 regularized...

Does Learning Require Memorization? A Short Tale about a Long Tail

State-of-the-art results on image recognition tasks are achieved using o...

Please sign up or login with your details

Forgot password? Click here to reset