arXiv ScienceSearch

arXiv subjects

Mark Spanier

Publications and source records attributed to Mark Spanier.

4 recordsLinked to original sources

Detection and Prevention Against Poisoning Attacks in Federated Learning

This paper proposes and investigates a new approach for detecting and preventing several different types of poisoning attacks from affecting a centralized Federated Learning model via average accuracy deviation detection (AADD). By comparing each client's accuracy to all clients' average accuracy, AADD detect clients with an accuracy deviation. The implementation is further able to blacklist clients that are considered poisoned, securing the global model from being affected by the poisoned nodes. The proposed implementation shows promising results in detecting poisoned clients and preventing the global model's accuracy from deteriorating.

cs.CR

Weighted uniform convergence of entire Grünwald operators on the real line

We consider weighted uniform convergence of entire analogues of the Grünwald operator on the real line. The main result deals with convergence of entire interpolations of exponential type $τ>0$ at zeros of Bessel functions in spaces with homogeneous weights. We discuss extensions to Grünwald operators from de Branges spaces.

math.CA

Extremal Signatures

Let $E= A - iB$ be a Hermite-Biehler entire function of exponential type $τ/2$ where $A$ and $B$ are real entire, and consider $dμ(x) = dx/|E(x)|^2$. We show that the sign of the product $A B$ is an extremal signature for the space of functions of exponential type $τ$ with respect to the norm of $L^1(μ)$. This allows us to find best approximations by entire functions of exponential type $τ$ in $L^1(μ)$-norm to certain special functions (e.g., the Gaussian and the Poisson kernel).

math.CA

Extremal functions with vanishing condition

We determine the optimal majorant $M^+$ and minorant $M^-$ of exponential type for the truncation of $x\mapsto (x^2+a^2)^{-1}$ with respect to general de Branges measures. We prove that \[ \int_\mathbb{R} (M^+ - M^-) |E(x)|^{-2}dx = \frac{1}{a^2 K(0,0)} \] where $K$ is the reproducing kernel for $\mathcal{H}(E)$. As an application we determine the optimal majorant and minorant for the Heaviside function that vanish at a fixed point $α= ia$ on the imaginary axis. We show that the difference of majorant and minorant has integral value $(πa - \tanh(πa))^{-1} πa$.

math.CA