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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Nowadays, systems based on machine learning (ML) are widely used in differentdomains. Given their popularity, ML models have become targets for variousattacks. As a result, research at the intersection of security/privacy and MLhas flourished. Typically such work has focused on individual types ofsecurity/privacy concerns and mitigations thereof. However, in real-lifedeployments, an ML model will need to be protected against several concernssimultaneously. A protection mechanism optimal for one security or privacyconcern may interact negatively with mechanisms intended to address otherconcerns. Despite its practical relevance, the potential for such conflicts hasnot been studied adequately. We first provide a framework for analyzing such"conflicting interactions". We then focus on systematically analyzing pairwiseinteractions between protection mechanisms for one concern, model and dataownership verification, with two other classes of ML protection mechanisms:differentially private training, and robustness against model evasion. We findthat several pairwise interactions result in conflicts. We explore potentialapproaches for avoiding such conflicts. First, we study the effect ofhyperparameter relaxations, finding that there is no sweet spot balancing theperformance of both protection mechanisms. Second, we explore if modifying onetype of protection mechanism (ownership verification) so as to decouple it fromfactors that may be impacted by a conflicting mechanism (differentially privatetraining or robustness to model evasion) can avoid conflict. We show that thisapproach can avoid the conflict between ownership verification mechanisms whencombined with differentially private training, but has no effect on robustnessto model evasion. Finally, we identify the gaps in the landscape of studyinginteractions between other types of ML protection mechanisms.

Conflicting Interactions Among Protection Mechanisms for Machine Learning Models