Quantum Differentially Private Sparse Regression Learning
AI Breakdown
Get a structured breakdown of this paper — what it's about, the core idea, and key takeaways for the field.
Abstract
The eligibility of various advanced quantum algorithms will be questioned if they can not guarantee privacy. To fill this knowledge gap, here we devise an efficient quantum differentially private (QDP) Lasso estimator to solve sparse regression tasks. Concretely, given <inline-formula> <tex-math notation="LaTeX">$N~d$ </tex-math></inline-formula>-dimensional data points with <inline-formula> <tex-math notation="LaTeX">$N\ll d$ </tex-math></inline-formula>, we first prove that the optimal classical and quantum non-private Lasso requires <inline-formula> <tex-math notation="LaTeX">$\Omega (N+d)$ </tex-math></inline-formula> and <inline-formula> <tex-math notation="LaTeX">$\Omega (\sqrt {N}+\sqrt {d})$ </tex-math></inline-formula> runtime, respectively. We next prove that the runtime cost of QDP Lasso is <italic>dimension independent</italic>, i.e., <inline-formula> <tex-math notation="LaTeX">$O(N^{5/2})$ </tex-math></inline-formula>, which implies that the QDP Lasso can be faster than both the optimal classical and quantum non-private Lasso. Last, we exhibit that the QDP Lasso attains a near-optimal utility bound <inline-formula> <tex-math notation="LaTeX">$\tilde {O}(N^{-2/3})$ </tex-math></inline-formula> with privacy guarantees and discuss the chance to realize it on near-term quantum chips with advantages.