Evaluating Multiple Guesses by an Adversary via a Tunable Loss Function
Abstract
We consider a problem of guessing, wherein an adversary is interested in knowing the value of the realization of a discrete random variable on observing another correlated random variable . The adversary can make multiple (say, ) guesses. The adversary's guessing strategy is assumed to minimize -loss, a class of tunable loss functions parameterized by . It has been shown before that this loss function captures well known loss functions including the exponential loss (), the log-loss () and the - loss (). We completely characterize the optimal adversarial strategy and the resulting expected -loss, thereby recovering known results for . We define an information leakage measure from the -guesses setup and derive a condition under which the leakage is unchanged from a single guess.
Cite
@article{arxiv.2108.08774,
title = {Evaluating Multiple Guesses by an Adversary via a Tunable Loss Function},
author = {Gowtham R. Kurri and Oliver Kosut and Lalitha Sankar},
journal= {arXiv preprint arXiv:2108.08774},
year = {2021}
}
Comments
6 pages