English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Differential Privacy (DP) is a key property to protect data and models fromintegrity attacks. In the Deep Learning (DL) field, it is commonly implementedthrough the Differentially Private Stochastic Gradient Descent (DP-SGD).However, when a model is shared or released, there is no way to check whetherit is differentially private, that is, it required to trust the model provider.This situation poses a problem when data privacy is mandatory, specially withcurrent data regulations, as the presence of DP can not be certificatedconsistently by any third party. Thus, we face the challenge of determiningwhether a DL model has been trained with DP, according to the title question:Can we infer the presence of Differential Privacy in Deep Learning models'weights? Since the DP-SGD significantly changes the training process of a DLmodel, we hypothesize that DP leaves an imprint in the weights of a DL model,which can be used to predict whether a model has been trained with DPregardless of its architecture and the training dataset. In this paper, wepropose to employ the imprint in model weights of using DP to infer thepresence of DP training in a DL model. To substantiate our hypothesis, wedeveloped an experimental methodology based on two datasets of weights of DLmodels, each with models with and without DP training and a meta-classifier toinfer whether DP was used in the training process of a DL model, by accessingits weights. We accomplish both, the removal of the requirement of a trustedmodel provider and a strong foundation for this interesting line of research.Thus, our contribution is an additional layer of security on top of the strictprivate requirements of DP training in DL models, towards to DL models.

Can we infer the presence of Differential Privacy in Deep Learning models' weights? Towards more secure Deep Learning