Researchers have demonstrated that adaptive attacks on linear sketches, which compress high-dimensional data, are fundamentally limited by a quadratic barrier. Even with a responder that knows all attack details and can adapt its strategy, the optimal privacy guarantee is $\widetilde{\Omega}(k^2)$, where $k$ is the sketch dimension. The study shows that for any hidden matrix $A$ and accuracy $\alpha$, an adaptive black-box attack can force a significant fraction of answers to have substantial relative error, clarifying the limits of defending linear representations. AI
RANK_REASON Academic paper detailing theoretical computer science research on data compression and privacy. [lever_c_demoted from research: ic=1 ai=0.4]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
