Additive Logistic Mechanism for Privacy-Preserving Self-Supervised Learning. (arXiv:2205.12430v1 [cs.LG])

We study the privacy risks that are associated with training a neural
network's weights with self-supervised learning algorithms. Through empirical
evidence, we show that the fine-tuning stage, in which the network weights are
updated with an informative and often private dataset, is vulnerable to privacy
attacks. To address the vulnerabilities, we design a post-training
privacy-protection algorithm that adds noise to the fine-tuned weights and
propose a novel differential privacy mechanism that samples noise from the
logistic distribution. Compared to the …

arxiv learning privacy self-supervised learning supervised learning