Reproduce “Occluded Oculus: Operationalizing Stylistic Obscurement” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Robert Dilworth
An ablation study of the TraceTarnish adversarial stylometry attack systematically evaluated its modular components, finding that character-level Injection techniques are most effective at confounding authorship attribution systems. The research demonstrates Injection's ability to consistently misclassify text authorship, offering a method to enhance digital privacy against sophisticated stylometric analysis.
Key insights
- Character-level injection techniques are significantly more effective at disrupting stylometric authorship attribution than semantic or stylistic modifications.
- Successful adversarial stylometry does not necessarily make text untraceable, but rather creates a distinct stylistic trace that allows texts to blend into a homogenous "crowd" of similarly modified texts, enhancing anonymity.
- The impact of adversarial techniques on authorship attribution is highly dependent on the specific linguistic features they perturb, with direct alterations to character structure proving most potent.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.